Budding Yeast Cell Cycle Analysis and Morphological Characterization by Automated Image Analysis

Size: px
Start display at page:

Download "Budding Yeast Cell Cycle Analysis and Morphological Characterization by Automated Image Analysis"

Transcription

1 Budding Yeast Cell Cycle Analysis and Morphological Characterization by Automated Image Analysis by Elizabeth Perley B.Sc., Massachusetts Institute of Technology, 21 Submitted to the Department of Electrical Engineering and Computer Science in Partial Fulfillment of the Requirements for the Degree of Master of Engineering in Electrical Engineering and Computer Science at the Massachusetts Institute of Technology May 211 MASSACHUSETVS INSTTUTE OF TECHL.O' y JUN [3unt Massachusetts Institute of Technology All rights reserved. ARCHIVES The author hereby grants to M.I.T. permission to reproduce and to distribute publicly paper and electronic copies of this thesis document in whole and in part in any medium now known or hereafter created. L IBRA R IES A u th or Department of Electrical Engineering and Computer Science May 2, 211 C ertified.by... Mark Bathe, Assistant Professor Thesis Supervisor A ccepted by Dr. Christopher J. Terman Chairman, Masters of Engineering Thesis Committee

2 Budding Yeast Cell Cycle Analysis and Morphological Characterization by Automated Image Analysis by Elizabeth Perley Submitted to the Department of Electrical Engineering and Computer Science in partial fulfillment of the requirements for the degree of Master of Engineering Abstract Budding yeast Saccharomyces cerevisiae is a standard model system for analyzing cellular response as it is related to the cell cycle. The analysis of yeast cell cycle is typically done visually or by using flow cytometry. The first of these methods is slow, while the second offers a limited amount of information about the cell's state. This thesis develops methods for automatically analyzing yeast cell morphology and yeast cell cycle using high content screening with a high-capacity automated imaging system. The images obtained using this method can also provide information about fluorescently labelled proteins, unlike flow cytometry, which can only measure overall fluorescent intensity. The information about yeast cell cycle stage and protein amount and localization can then be connected in order to develop a model of yeast cellular response to DNA damage. Thesis supervisor: Mark Bathe Supervisor title: Assistant Professor

3 Table of Contents Abstract Table of Contents List of Figures List of Tables 1 Introduction 2 Related work and Background 2.1 Background Yeast cell cycle stage and response to DNA damage Yeast Cell Cycle Analysis and Morphological Characterization by Multispectral Imaging Flow Cytometry Current cell detection and classification software C ellprofiler Overview Image Processing 4.1 Im aging Bright field images vs. Concanavalin A images Cell detection and segmentation Edge detection with watershedding of bright field images Thresholding using Concanavalin A Voronoi-based segmentation using CellProfiler Yeast-specific cell detection and segmentation Nucleus detection and segmentation Discussion Cell Cycle Stage Classification Creation of a training set Feature selection

4 5.3 Feature calculation Basic cell features Bud detection Feature validation Using the training set Using cells arrested in stages of cell cycle 5.5 Classification using neural nets Creation of the neural net Net performance on training set Net performance on additional data... 6 Conclusions Appendices 7.A Edge detection and watershedding code B Concanavalin A thresholding code C Nucleus detection code D Feature calculation code E Bud detection code References

5 List of Figures Bright field image of budding yeast cells Bright field image of budding yeast cells Comparison of bright field and ConA images Edge detection and watershedding to segment cells... Distance transform of cells Edge detection and watershedding on ConA images... CellProfiler Cell Segmentation Polar plot of yeast cells for segmentation Yeast-specific cell detection algorithm to detect cells..... Budding yeast cell cycle stages. Adapted from Calvert, et al. Cell undergoing bud detection algorithm Cell area feature distribution Average nuclear intensity distributions Overall nuclear intensity distributions Bud size distributions Bud/mother cell ratio distributions Cell area - Arrested Cells vs. Training set Comparison [2] Average nuclear intensity - Arrested Cells vs. Training set Comparison Bud size - Arrested Cells vs. Training set Comparison Neural net used for classification Cell area feature distribution Average nuclear intensity distributions Overall nuclear intensity distributions Bud size distributions Bud/mother cell ratio distributions

6 List of Tables 1 Feature list for classification Differentiation between G 1 and G 2 using a feature subset Differentiation between G 1, S and G 2 /M using a feature subset Estimated cell diameter Neural net performance on training set Neural net performance on all data Estimated cell diameter Neural net performance on arrested cells

7 1 Introduction In recent years there has been a significant amount of development in the area of high content (HC) imaging. This technique attempts to combine high throughput with high resolution imaging to provide statistics with a large number of samples, and images which provide accurate measurements. A number of platforms and a several pieces of software are now available to acquire and analyze images from HC screens. However, most of these were developed specifically for mammalian cells, which are large and grow in single layers, and therefore make assumptions about the types of images that can be used as input lack the ability to deal with other types of cells. One type of cell that these platforms are often not capable of dealing with is the budding yeast (Saccharomyces cerevisiae) cell. Budding yeast cells are one of the standard model systems in biology. Their cell-cycle has been well characterized in terms of cellular morphology as well as in the proteins involved, and there is a large amount of biological knowledge that exists. This makes them an excellent choice for developing models. However, these cells do not grow in a single layer, and they are quite small compared to mammalian cells, so in order to use HC imaging, a new tool for analysis must be used. The first part of this thesis describes methods to analyze HC images of budding yeast cells. First, cell outlines must be determined from the images of cells. Each well on a plate contains hundreds of cells and requires multiple images to be taken, each of which contains tens of cells. These cells must be correctly identified, and accurately segmented and outlined to get the correct cell shape, as well as to allow for the correct calculation of protein levels and localization. This problem poses several challenges: The images are taken automatically, and as a result are not always completely focused. They also often have uneven illumination, and contain other particles or out of focus cells, which contribute to noise and misidentification of cells. Budding yeast cells also have very thick cell walls, making the decision to find the

8 Figure 1: Bright field image of budding yeast cells "correct" cell border a difficult one. This can be seen in Figure 1. It is unclear if it is at the outer cell border, or the inner cell border, or somewhere in between. Several different approaches for cell segmentation were investigated, and the best selected. The second part of this thesis describes a way to use these image analysis methods to develop a model of how budding yeast cells respond to DNA damage. When a cell's DNA becomes damaged, the cell must respond in some way in order to repair its DNA and continue its progress in the cell cycle. Many genes and pathways have been found that are responsive to DNA damage in budding yeast, using techniques such as bulk transcriptional profiling and genomic phenotyping assays. However, little is known about the mechanism by which these changes occur, and how protein levels and localization with in the cell are affected by damage. Previously, flow cytometry was used to study this response. However, this method only allows overall protein and DNA levels to be measured. It has been shown that the response of budding yeast cells is dependent on their current stage of the cell cycle, which cannot be seen using flow cytometry. Although it is possible to arrest cells at specific stages of the cell

9 cycle with drugs, these may introduce artifacts. Since HC imaging allows cell morphology to be examined and cell cycle determined, this approach offers a significant advantage and it can provide a more detailed picture of the response of budding yeast. The paper describes a machine learning approach to classifying yeast cells into their cell cycle stages. Once the cell cycle stages can be accurately identified from cell shape and nuclear shape, DNA content and position, then asynchronous populations of yeast cells can be characterized. Algorithms to find the bud of a cell and calculate morphological features of the cells were developed. Using these features, neural nets were tested as ways to classify these cells in a supervised learning approach.

10 2 Related work and Background 2.1 Background Yeast cell cycle stage and response to DNA damage Previous work has been done to determine the cellular response of budding yeast to DNA damaging agents by Jelinsky, et al.[1] In their paper, budding yeast cells were exposed to various carcinogenic alkylating agents and oxidizing agents, as well as ionizing radiation, and it was found that this exposure modulates transcript levels for over one third of the genes of the yeast cells. What was particularly relevant about these results was the finding that for one of the carcinogenic agents, MMS, the response is dramatically affected by the cell's position in the cell cycle at the time that it is exposed to the agent. Cultures of log-phase budding yeast cells were arrested in G 1 by a-factor, in S phase by hydroxyurea, and in G 2 by nocodazole. These cells were then allowed to grow for three days before each group of cells was split in half and MMS added to one of the two halves in each phase. In order to investigate the cellular response to the addition of MMS, GeneChip hybridizations were used. A set of arrays was used contianing probes for 6,218 yeast ORFs and then analyzed using the GeneChip analysis suite. Then, any gene which showed a change of 3.- fold or more in at least one of the experimental conditions was further examined. The result was that there were 693 such genes which were responsive to treatment with MMS. These were shown to be certainly responsive, with none of the error bars for treated and untreated cells coming close to overlapping. Initially, when asynchronous log-phase cultures were treated with MMS, many genes were scored only as weakly responsive. However, treating yeast cells which had been arrested in

11 the stages of the cell cycle caused many more genes to be shown as clearly responsive to the treatment. Of the genes that were responsive, 199 were responsive only in G 1, 84 were responsive only in S phase, 94 were responsive only in G 2, and 229 were responsive only in stationary phase. Prior to these experiments, fewer than 2% of the genes examined had been shown to have cell-cycle dependent expression. These results indicate that, in order to fully understand the budding yeast response to damage, each cell's stage in the cell cycle must be taken into account Yeast Cell Cycle Analysis and Morphological Characterization by Multispectral Imaging Flow Cytometry Yeast cell cycle analysis has previously been done using multispectral imaging flow cytometry (MIFC) in a computational approach. [2] In a paper by Calvert, et al., an improvement on the traditional yeast cell cycle analysis using flow cytometry is developed. MIFC offers multiple parameters for morphological analysis, allowing for the calculation of a novel feature, bud length, which is used to observe the change in morphological phenotypes between wild-type yeast cells, and those which overexpress NAP1, which causes an elongated bud phenotype. An imaging flow cytometer was used, which allows the simultaneous detection of bright field, dark field, and four fluorescent channels. The images from MIFC were then used to visually assign cells to one of the three stages of the cell cycle: G 1, S, and G 2 /M. They were distinguished as cells in G 1 being those that have a single nucleus and no bud, S being cells with a single nucleus and a visible bud, and G 2 /M being those with elongated or divided nuclei and a large bud. The yeast cell cycle was determined using a combination of DNA intensity and nuclear morphology. Cells were split into ones that contained IN DNA content and round nuclei, 2N DNA content and round nuclei, and 2N DNA content and elongated or divided nuclei,

12 and these three groups were labelled as G 1, S, and G 2, respectively. 1 cells from each stage of the cell cycle were then visually identified and labelled in order to validate this method of determining cell cycle, which was 99% accurate. Further testing of this method of cell cycle analysis was performed, and results from visual analysis of morphology and from this method of separation were similar, and far more accurate that analysis with standard flow cytometry using a cell cycle modelling program. The bright field images were analyzed automatically to calculate the bud length feature. To do this, an object mask for the cell was calculated by separating the cell from its background using pixel intensity in the bright field channel. This mask was then eroded by 3 pixels. Then, the bud length was calculated by subtracting the maximum thickness of the cell calculated from MIFC, which is assumed to be the diameter of the mother cell, from the total length of the cell as determined from the object mask. Relative bud length was calculated to be the ratio of the minimum thickness of the cell calculated from MIFC, and the width of the bud. The cell aspect ratio was calculated as the ratio between the total cell length and the width. Then, any cell which had a relative bud length larger than 1.5, and an aspect ratio of less than.5 was considered to be a cell with an elongated bud. The NAP1 strains and wild-type strains were then analyzed for differences in bud length as well as cellular shape. It was shown that MIFC could accurately distinguish between the elongated bud phenotype and a normal bud phenotype. This approach shows that it is possible to determine cell cycle stage with MIFC using simple computational methods. In this case, a simple scatter plot was used to separate the cells between G 1, S and G 2, using DNA content and nuclear morphology. It also shows that morphological features can be generated using cell shape and size.

13 2.2 Current cell detection and classification software CellProfiler CellProfiler aims to perform automatic cell image analysis for a variety of phenotypes of cells from different organisms. [4] It is useful for measuring a number of cell features, including cell count, cell size, cell cycle distribution, organelle number and size, cell shape, cell texture, and the levels and localization of proteins and phosphoproteins. The motivation behind the creation of CellProfiler is to provide quantitative image analysis, as opposed to human-scored image analysis, which is qualitative and usually classifies samples only as hits or misses. It also allows the processing of data at a much quicker rate, and the creators consider cell image analysis to be one of the greatest remaining challenges in screening. The software system consists of many already- developed methods for many cell types and assays. It uses the concept of a pipeline of these individual modules, where each module processes the images in some manner, and the modules are placed in a sequential order. The standard order for this processing first contains image processing, then object identification, and finally measurement. According to the developers of CellProfiler, the most challenging step in image analysis is object identification. It is also one of the most important steps, since the accuracy of object detection determines the accuracy of the resulting cell measurements. The standard approach to object identification in CellProfiler is to first identify primary objects, which are often nuclei identified from DNA-stained images. Once these primary objects have been identified, they are used to help identify and segment secondary objects, such as cells. The identification first of primary objects helps to distinguish between different secondary objects, since they are often clumped. In the software, first the clumped objects are recognized and separated, then the dividing lines between them are found. Some of these algorithms for object identification are discussed later in subsection 4.2. One of these was specifically

14 developed for CellProfiler, which was an improved Propagate algorithm and allowed cell segmentation for some phenotypes to be performed which had never been possible previously. The main testing of CellProfiler was performed on Drosophila KC167 cells because they are particularly challenging to identify using automated image analysis. It was also tested on many different types of human cells, and has been shown to work well on mouse and rat cells as well. One of the main goals of CellProfiler was to be able to identify a variety of phenotypes and to be flexible, modular, and to make setting up an analysis feasible for non-programmers. This means that it does not have many modules for specific cell types, but rather more general ones that are applicable to many different types, which does make it good for many applications. However, for specific applications, or for cell types that it wasn't designed for, CellProfiler is not always a good fit.

15 ... 3 Overview The approach to examine budding yeast cellular response to DNA damage is outlined in Figure 2. First, images of cells are acquired using a Cellomics high-content automated imaging system. They then undergo image processing, which involves cell detection and segmentation, as well as nucleus detection. As described in the CellProfiler approach, this step is the most important because it determines the accuracy of all other measurements. Segment cells A Segment nuclei Measure features of cells (Size, intensity, bud size, etc) Classify cells according to stage of cell cycle Calculate statistical metrics to determine quality of features and classification Figure 2: Bright field image of budding yeast cells Once the cells have been detected, then features of these cells are measured, such as cell size, shape, and nuclear size, shape, and intensity. These features are then used as input to a supervised learning system in order to classify cells as being in one of the three cell-cycle stages: G1, S, or G2/M. Finally, the features and the classification system are validated and tested. Once cell cycle can be accurately determined, it is possible to measure other features of the cell from fluorescent channels and combine these measurements with the cell cycle information to learn more about the yeast cell response to DNA damage.

16 4 Image Processing 4.1 Imaging The yeast library which was used was developed at the University of California San Francisco, and is now commercially available. It consists of yeast strains in which individual proteins are expressed with C-terminal GFP tags from endogenous promoters. The cells were fixed and stained with DAPI for visualizing the DNA, and with Concanavalin A for visualizing the cell walls. These cells were then imaged on 96-well plates using a Cellomics system from Thermofisher Scientific. Images were acquired in three fluorescent channels (for DAPI, GFP, and Concanavalin A) and in the bright field channel to allow ready visualization of cell contours Bright field images vs. Concanavalin A images The standard way of viewing cell morphology is through the use of bright field images. These images show the basic cell outline of the budding yeast cells. However, these bright field images might not provide the best data for a high content screening approach for the following reasons: " It is difficult even for a human to determine what the actual boundary of the cell is due to the thickness of the cell wall in the bright field images " Cell buds are difficult to see due to the thickness of the cell wall, especially when they are small. * Bright field images show not only cells, but also other particles in the well, and out of focus cells, which must be removed for further analysis of the cell images

17 Despite these problems, bright field is a type of image worth exploring for looking at cells because it requires no additional stains or steps in cell preparation. It also doesn't require the use of a GFP channel on the Cellomics imaging platform, which only has a limited number of channels. Another choice for viewing cell morphology is to use cells which have been stained with fluorescent-conjugated Concanavalin A (ConA). ConA is a lectin which combines with proteins on the budding yeast cell walls. ConA provides a much clearer image of yeast cell morphology for the following reasons: " Cell outlines are well defined with no ambiguity " Even small buds can be easily seen " There is a minimal amount of noise, with no extra particles or out-of-focus cells, due to the fact that this is a fluorescent image However, this requires the use of an additional channel in the microscope, which is only capable of taking pictures in 4 channels. Although in this paper not all channels are needed, it is desirable to have additional channels available so that multiple molecules in the cell may be observed. The other drawback to using ConA images is the fact that the definitions between cells isn't quite as clear. While it is quite easy to tell which parts of the images are cells and which aren't, it can be difficult to distinguish between different cells in a cluster. This is due to the fact that there is no actual outline of the cell, and the entire cell is stained with ConA. However, use of a more dilute sample can alleviate this problem. In subsubsection 4.1.1, the comparison between bright field and ConA images is clear. ConA images provide a much clearer view of the cells. Although some groups have had success with segmenting cells using bright field images alone [13], ConA images were chosen

18 Figure 3: Comparison of bright field (left) and ConA (right) images of budding yeast cells after the investigation of the use of both types of images. Bud size is an easy way to roughly determine cell cycle stage, so the ability to calculate precise cell outlines was a major deciding factor in the choice of images. It is also important that the amount of protein in the cell can be determined. If the correct cell outlines are not found, then the amount of protein can be overestimated or underestimated. 4.2 Cell detection and segmentation The cells in any image must first be separated from their background before further analysis can be done. While for many applications, only a cell count or a rough estimate of the cell shape and size is necessary, an accurate border is required in order to detect the bud and compute levels of fluorescence correctly.

19 4.2.1 Edge detection with watershedding of bright field images There are three major challenges in detecting and segmenting cells. One of these is that the bright field images are the preferred choice for cell detection. However, the cell cannot simply be thresholded out in these images due to a low level of contrast between cells and the background. Another such challenge is the presence of non-cellular objects within the bright field images. Each object that is detected must be a cell, so any object that might be mistaken for a cell must be identified and removed. The last challenge is the fact that the cells are often grouped very closely together, and the cell detection algorithm must be able to separate these clusters. Edge detection One way to detect the cells in the low-contrast bright field images is to perform edge detection. Edge detection algorithms calculate the gradient across the image and then use a cutoff threshold to determine which values are a part of an edge, and which are not. This procedure uses the Canny edge detection algorithm. [3] This algorithm is designed to mark each edge only once, detect edges as close as possible to the actual edges in the image, and to not be affected by noise in the original image. The cell detection algorithm is outlined as follows, and code be found in appendix 7.A: 1. Apply a Gaussian filter to reduce the amount of noise in the original image 2. Use the Canny algorithm to detect edges in the image. 3. Dilate the detected edges with linear elements. This fills in small gaps in the edges, which occur due to the fact that the image is very low-contrast, making it difficult to detect a full outline of each cell. 4. Fill in the interiors of the cells by filling in every closed outline in the image.

20 5. Erode with a circular element to smoothen the cells. 6. Remove objects that are too small to be cells. Although the Canny edge detection algorithm has a noise removal step in it, the first step of the cell detection applies a Gaussian blur for further noise removal. This causes the edge detection to detect fewer incorrect small edges, since the cells are large enough such that this step does not affect their edges significantly. The MATLAB canny function can takes an argument that specifies the sensitivity threshold. In the code, a sensitivity value of.3 was chosen, which was also chosen to minimize the number of incorrect small edges while still detecting the main cell outlines. Once edges have been detected, there are often gaps in the full cell outlines. These must be removed so that a cell mask can be created by filling in the outlines. Dilating the edges with a small horizontal and vertical linear element fills in the majority of these small gaps. After the cells have been filled in, the cell mask which has been created is slightly larger than the cell. This is partially due to the fact that the edges were dilated, but also due to the fact that the cell wall is quite thick in the bright field images. For any particular cell, two edges are detected: the outside edge of the cell wall, and the inside edge of the cell wall. The cell border is actually somewhere in between these two lines, so the cell must be eroded a small amount. This erosion step can also help to smoothen the jagged edges that can be created due to the edge detection. A circular element is used for the erosion to maintain as much of the original cell shape as possible. The erosion also removes any objects which are too thin in one dimension to possibly be a cell. Although the last erosion step is necessary, it removes quite a bit of detail from the cell outlines. While it smooths the jagged edges, it also can remove small buds from the cell. It also erodes the cell shape such that it is not always clear where the bud of the cell is exactly.

21 Figure 4: Using edge detection and watershedding to segment cells Watershedding After cell detection has been performed, the cells must still be segmented, since clusters of cells still remain. One standard algorithm which is used for cell segmentation is watershedding. [6] The basic concept behind watershedding is that the image is treated as a height field, and then the pixels are separated based on which minimum a drop of water would flow to if it were placed on that pixel in the height field. In order to perform watershedding, first a distance transform is calculated on the cell mask obtained from the cell detection process. This involves calculating the distance from any pixel which is part of a cell to the closest pixel which is not part of a cell. So, pixels which are at the center of a cell have the highest values. An example of this can be seen in Figure The values are then negated, and watershedding is performed. This means that any pixel which is considered to be part of a cell is assigned to the local minimum to which a drop of water would flow to if it were placed on that pixel in the negated distance transform. In Figure these local minima would be the brightest pixels in the distance

22 Figure 5: Comparison of cell mask (left) with its distance transform (right) transform. However, watershedding is not an ideal method for cell segmentation. It can be practical because it is quick and can give a reasonable estimate of the number of cells in an image as well as the approximate area of the cells. However, a single noisy pixel can affect the grouping of an entire cluster of pixels by creating a gap between the group and a diferent minimum. The cell borders which are finally detected are also only somewhat related to the original image, meaning that when the cells in a cluster are segmented, the line chosen is based on the distance transform, rather than any edges that were present in the original image. Then, by the time the watershedding is performed, the cell outlines have already been manipulated so much that the distance transform is not a good measure of where the cells lie. Even on images which are not noisy and segmentation should be easy, such as the ConA images, edge detection with watershedding does not perform well. This can be seen in Figure 6, where this same algorithm was run on ConA images of cells. Although the initial segmentation of the cells was trivial, the subsequent steps caused much of the clarity of the cell outlines and shapes to be lost, resulting in a final cell mask which has lost a great deal of information.

23 Figure 6: Comparison of original ConA image (left) with its cell outlines after edge detection (middle) with the final output after smoothing and watershedding (right) Thresholding using Concanavalin A Images of yeast cells taken using Concanavalin A stain in a fluorescent channel provides a much better source image to work with than bright field images. The cell outlines are clear, and since the images are high-contrast, thresholding the cells from their background is possible. This makes cell detection a trivial task. The cell detection algorithm is outlined below, and its code can be found in appendix 7.B: 1. Use a Gaussian filter to remove noise in the image 2. Adjust the contrast of the image so that the brightest pixels are as bright as possible and the darkest are as dark as possible. This helps to prepare the image so that it can be thresholded most easily. 3. Perform morphological closing on the image, to remove uneven brightness in the cells. [6] This occurs due to the fact that the edges of the cells appear brighter in the ConA images, since ConA is a cell wall stain. Morphological closing helps to remove gaps in the middle of the cell. 4. Threshold out the cells using an automatically calculated threshold value

24 5. Fill in the interiors of the cells by filling in every closed outline in the image. 6. Fill in the gaps in any outlines of the cells using linear line elements, and then fill in any new holes created. 7. Remove objects which are too small or too large to be cells The first step of the cell detection applies a Gaussian blur of size 3 x 3 for further noise removal. This smooths the cells to allow smooth cell borders to be detected once the image is thresholded. Then the values of the image are scaled such that they take on the full range of possible pixel values. Spreading the values out in this way allows the image to be thresholded more easily. The morphological closing is performed to fill in gaps in the middle of the cells and to even out some of the cell brightness. Though the interiors of the cells will be later filled in, this is done as a preventative measure to also ensure that there are not gaps at the edge of the cells which cannot get filled in. This image is then turned into a black and white cell mask using an automatically calculated threshold value. The standard MATLAB greythresh method is used. This method implements Otsu's method, which chooses the threshold to minimize the intraclass variance of the black and white pixels. [16] This function is used to calculate a starting point for a cutoff threshold, and then this is scaled by a factor of.8 to adjust it for the image set being used. After the cells are filled in, there are still some remaining portions of cell outlines which have not been filled in. The image is dilated with small linear elements using many different angles to fill in the largest number of gaps possible. Any new cell interiors which have been created are then filled in using MATLAB's imf ill method, which fills in any background pixels which cannot be reached from the edge of the image. However, cell segmentation still presents a problem. Watershedding loses information about cell shape as shown in the previous section. This problem can be dealt with by using a sample of cells which is dilute enough such that there are few clusters of cells, which are

25 then not included in further data analysis (they get removed in the last step of the cell detection algorithm). It also might be possible to segment these cells using an algorithm such as that described in the next sections. This problem ended up being beyond the scope of the thesis, but would be good for further investigation Voronoi-based segmentation using CellProfiler The CellProfiler software discussed in subsubsection uses a novel method of segmenting cells using Voronoi regions. [12] This approach was designed to overcome the limitations of watershedding that cause it to be a fragile algorithm for segmenting cells. It does so by comparing neighborhoods of pixels rather than individual pixels to avoid the issue of a single noisy pixel affecting the segmentation of a group of pixels. It also tries to segment cells based on the borders found in the original image but includes a regularization factor to provide reasonable behavior when there is not a strong edge between two cells. Unlike the other algorithms which relied only on the image of the cell morphology to detect and segment cells, this approach relies on an image of the nuclei as well. It considers these nuclei to be seed regions, and proceeds to find a cell for each nucleus detected. The CellProfiler platform provides an implementation of this algorithm, and the image of detected nuclei from subsection 4.3 was used to provide the seed regions for the algorithm. Both bright field and ConA images were used as potential source images for which to segment the cells. The results of two representative runs of the algorithm on these images are shown in Figure 7. In the runs with bright field images, the noisiness of the original image caused the detected cell borders to also be extremely noisy and jagged, often extending far beyond the actual cell border. The ConA images provided smoother borders for the most part, with some extreme missegmentations, like putting one cell inside another cell. The ConA images

26 Figure 7: Output of the CellProfiler segmentation algorithm run with the bright field (left) and ConA (right) images as input superimposed on the original bright field image.. Nuclei used as seed regions are outlined in green and detected cell borders are shown in red. also often had borders which did not align with cell borders, or which extended beyond cell borders. Results While this algorithm initially appeared promising, results were not as good as those obtained from thresholding. One possible explanation for the poor quality of cell detection and segmentation was the fact that the cells being segmented are yeast cells, while CellProfiler was designed mostly for mammalian cells where most cells are sharing borders with other cells. Yeast cell morphology is significantly different from that of other cells. They are smaller, so small errors in borders affect overall accuracy more, and they are not usually touching on all sides like mammalian cell do, though there are occasional clusters. However, it would be possible in future work with budding yeast cells to adapt this algorithm for their specific cell morphology. It did perform well in determining which objects were actually composed of more than one cell and approximating where the borders might be between them.

27 4.2.4 Yeast-specific cell detection and segmentation It is appropriate to look at a cell detection algorithm tailored specifically to yeast cells given the fact that the more general algorithm in the previous section did not perform well due to cell morphology differences. Although there are not many yeast-specific segmentation approaches, one group developed an approach that performed well. [13] This approach uses a bright field includes a scheme for cell detection as well as segmentation, both of which rely on the computation of a gradient image. This gradient image helps to eliminate problems of uneven illumination. The segmentation part of the algorithm relies on the detection of candidate cell centers in a cluster of cells and then the use of a polar plot of the cell to find the best cell contours. This approach shouldn't suffer from the same problems as had been described in subsubsection It chooses the best cell borders using dynamic programming, which optimizes the choice of a cell border by looking at the whole cell. It avoids the problem of having jagged cell borders by the nature of the dynamic programming approach, and it detects the cells initially by using a method that models noise in the image to eliminate it as a source for errors. The cell detection process can be outlined as follows: 1. Compute the gradient image from the original bright field image using Prewitt's method. [7] 2. Find the threshold at which to determine which gradient values are part of the cells and which are noise. This is calculated by fitting the gradient values to a distribution and removing those below a specific value (described in detail below). 3. Fill in remaining holes in the cells.

28 4. Perform a morphological opening with a small circular element to remove small structures due to noise. The output of this algorithm should be a set of cells ready to be segmented. The gradient image is calculated by filtering the image with two masks, one for each direction, to enhance differences in both directions, and then letting the final gradient image be the magnitude of these two filtered images. Then, any pixel with a value above a threshold, which is defined as # in the original paper, is assigned to be a part of the foreground. In order to calculate /, the gradient values below the median are fitted to a Rayleigh distribution function, where there is a parameter o which is varied until the best fit is found. Then, # = 7.5o-, which corresponds to designating pixels with gradient values more than 6 standard deviations larger than the mean of the estimated distribution of background pixels as a part of the foreground. This scheme relies on the assumption that the distribution of noise in the background regions of the image are approximately normally distributed. Once the cells have been detected from their backgrounds, the cells must then be segmented. The segmentation algorithm is outlined below: 1. Find candidate cell centers from the segmented image. 2. Create a polar plot of the cell from each candidate cell center. 3. Use dynamic programming with global constraints to choose an an optimal path from left to right on the polar plot. The original plot is repeated three times and the final chosen path is taken from the center repetition of the plot to ensure that the chosen path is closed. There are multiple ways to choose candidate cell centers depending on how clustered the cells are. If the clusters are no larger than 3 or 4 cells in each cluster, then a simple distance

29 a Figure 8: An example polar plot of two yeast cells calculated from the gradient image. This would have the dynamic programming algorithm run on it to find an optimal path transform on the cell mask can be used, and the local maxima are considered to be candidate cell centers, which is a good choice for this dataset. The polar plot from the candidate cell centers is then created by sampling rays outward at 3 equally spaced radial points. Then using this plot a cell contour can be extracted. The dynamic programming scheme uses constraints to ensure convexity for the majority of the cell, so it penalizes transitions in the polar plot which correspond to right turns. It does ensure that the extracted contour is closed by going around the cell three times. This algorithm should perform well even when the calculated candidate cell center is quite off-center. Results This approach relied on the ability to segment the cells by calculating the gradient image and then choosing a cutoff value from this image to determine which parts of the image were edges. However, it did not detect cell edges such that the cells could be filled in in the way that the algorithm describes. The cells were not even close to having closed outlines. To test if it was simply a problem of fitting the data correctly, the best value for # was chosen

30 manually for several images. However, the output was still not ideal. There was still the problem of the two detected edges for each cell wall as was discussed in earlier approaches to cell detection. It is difficult to know which cell edge to choose. Then, with the choice of #, it is not possible to guarantee that only one set of edges is included in the thresholded images, or that both sets of edges are included. Some partial edges are included, and in some cases, an entire outer edge is thresholded out. Finally, any partial outer edges which were detected are then removed with the morphological opening. This leads to a great deal of inconsistency in the final detected cells: some include the outer edges, and some don't. In cells with small buds, the buds are often lost as can be seen in the figure. Figure 9: Yeast-specific cell detection algorithm to detect cells. A threshold value # was chosen manually for this set of images This poor performance is most likely related to some of the assumptions made in the original paper about the distribution of noise in the image, and how it can be fit to a distribution. Though it claims that these assumptions are "crude", it says that they still allow the algorithm to perform well. This was not able to be reproduced, perhaps even due to some difference in the distribution of pixel values in the original data. Then, the second part of the scheme to segment cells was not implemented as a result of the failure of the first part. Though it could have been attempted using cells detected using other methods, it wasn't a worthwhile endeavor given the fact that the first part completely failed. The paper did admit that the only incorrectly detected cell contours were buds, which are important in this case. Though there was a proposed solution to this problem, it was

31 not tested. 4.3 Nucleus detection and segmentation The nuclei of the cells must also be detected and segmented in addition to the cells themselves. The cells, which were stained with ConA, were also stained with DAPI, a fluorescent stain which binds to A-T rich regions of DNA. This problem is relatively straightforward in comparison to that of segmenting cells, since the nuclei are never touching each other and clustered. The algorithm to detect nuclei is outlined below, and the code can be found in appendix 7.C 1. Use morphological opening to calculate the background of the image 2. Subtract this calculated background from the original image 3. Threshold out the nuclei to get a nuclear mask 4. Remove any objects which are too big to be nuclei 5. Dilate the nuclear mask since this thresholding tends to underestimate nuclear boundaries First the background of the image is calculated using morphological opening with a disk of radius 6. This removes any objects in the image which are smaller or thinner than the disk, which effectively removes all of the nuclei in the image. This then leaves only the background noise of the image, and any uneven illumination. This background calculation is important because in examining the nuclei, the intensity of the stain is important and can indicate the amount of DNA in the nucleus. Removal of the background ensures consistent intensity calculations. This step also helps to deal with one of the problems that can arise

32 with the DAPI staining: occasionally the entire cell will get stained with DAPI and the entire cell will fluoresce, or a much larger part of the cell will get stained with DAPI. These regions, which are larger than nuclei will, for the most part, be included in the background. Then, they will be removed in the next step when the background is subtracted from the original image. Once the background has been removed, the nuclei are thresholded out to create a nuclear mask. Then, any objects which are larger than 1 pixels are removed from the mask. This represents a nucleus with a radius of 6 pixels, which is the same as the disk size that was used for the morphological opening. Any nucleus should be much smaller than this. This step ensures that any nuclei which had problems with the DAPI staining are not included in the mask. In this nuclear detection code the DAPI channel is modified to remove any objects which are not nuclei, and to remove any background noise. In the last step, the nuclear mask is then dilated to make sure that the entire nucleus is included in the final modified image. It is not as important to get the nuclear outline perfect as it is to ensure that the entire nucleus is included in the mask, since the overall intensity of the nucleus is important, so the image is dilated several times with a disk of radius 1. In later steps in the workflow, the DAPI images will be combined with the segmented cells, and any nuclei which are not contained within cells will not be included in calculations. While it would be possible to start with a cell mask and then to look for a nucleus within the mask by performing operations locally, it is more simple and quite effective to do the operation for the entire image and apply the mask later. This order of operations also allows for the background intensity correction to occur for the entire image.

33 4.4 Discussion An ideal cell detection and segmentation algorithm would be able to identify only the parts of an image which are cells using smooth contours, and then correctly segment them. It would not leave out the contours of any buds, since those are necessary for cell cycle stage classification. It would also be precise in the detection of cell outlines, since this is necessary for calculating GFP intensity in the investigation into the response to DNA damage. An ideal algorithm would be able to segment even large clusters of cells so that any type of data could be used, with either dense cells or with sparse cells. Most importantly, the algorithm would be consistent in the way it detects and segments cells. The only approach that satisfied the majority of these requirements was that of using the ConA images with thresholding. This approach was selected because of its consistency and simplicity. The use of the ConA images completely eliminates the problem of cell wall thickness creating multiple edges, and thresholding is a reliable method of detection. Its drawback is that the images of cells must not contain large clusters, since these cannot be handled and will be ignored. This means that sparse images with dilute cells must be used. It might be possible to combine the ConA thresholding approach with one of those which performs segmentation by modifying both. However, it is possible that more complicated approaches such as that from subsubsection might not perform better in this context given the emphasis placed on bud detection. Once the rest of the methods are completely developed and tested, it will be possible to work on the cell detection and segmentation part of the workflow further to allow a wider variety of data to be able to be used as input.

204 Part 3.3 SUMMARY INTRODUCTION

204 Part 3.3 SUMMARY INTRODUCTION 204 Part 3.3 Chapter # METHODOLOGY FOR BUILDING OF COMPLEX WORKFLOWS WITH PROSTAK PACKAGE AND ISIMBIOS Matveeva A. *, Kozlov K., Samsonova M. Department of Computational Biology, Center for Advanced Studies,

More information

An automated image processing routine for segmentation of cell cytoplasms in high-resolution autofluorescence images

An automated image processing routine for segmentation of cell cytoplasms in high-resolution autofluorescence images An automated image processing routine for segmentation of cell cytoplasms in high-resolution autofluorescence images Alex J. Walsh a, Melissa C. Skala *a a Department of Biomedical Engineering, Vanderbilt

More information

ChIP-seq and RNA-seq. Farhat Habib

ChIP-seq and RNA-seq. Farhat Habib ChIP-seq and RNA-seq Farhat Habib fhabib@iiserpune.ac.in Biological Goals Learn how genomes encode the diverse patterns of gene expression that define each cell type and state. Protein-DNA interactions

More information

CHAPTER-6 HISTOGRAM AND MORPHOLOGY BASED PAP SMEAR IMAGE SEGMENTATION

CHAPTER-6 HISTOGRAM AND MORPHOLOGY BASED PAP SMEAR IMAGE SEGMENTATION CHAPTER-6 HISTOGRAM AND MORPHOLOGY BASED PAP SMEAR IMAGE SEGMENTATION 6.1 Introduction to automated cell image segmentation The automated detection and segmentation of cell nuclei in Pap smear images is

More information

Comparative Genomic Hybridization

Comparative Genomic Hybridization Comparative Genomic Hybridization Srikesh G. Arunajadai Division of Biostatistics University of California Berkeley PH 296 Presentation Fall 2002 December 9 th 2002 OUTLINE CGH Introduction Methodology,

More information

The Nuclear Area Factor (NAF): a measure for cell apoptosis using microscopy and image analysis

The Nuclear Area Factor (NAF): a measure for cell apoptosis using microscopy and image analysis The Nuclear Area Factor (NAF): a measure for cell apoptosis using microscopy and image analysis Mark A. DeCoster Department of Biomedical Engineering and Institute for Micromanufacturing, Louisiana Tech

More information

ChIP-seq and RNA-seq

ChIP-seq and RNA-seq ChIP-seq and RNA-seq Biological Goals Learn how genomes encode the diverse patterns of gene expression that define each cell type and state. Protein-DNA interactions (ChIPchromatin immunoprecipitation)

More information

Amnis ImageStream : Technical Reports & Applications

Amnis ImageStream : Technical Reports & Applications Amnis ImageStream : Technical Reports & Applications ImageStream : Flow Cytometry and Microscopy in a Single Platform The ImageStream achieves true multispectral Imaging in Flow by combining microscopy

More information

Brightfield and Fluorescence Imaging using 3D PrimeSurface Ultra-Low Attachment Microplates

Brightfield and Fluorescence Imaging using 3D PrimeSurface Ultra-Low Attachment Microplates A p p l i c a t i o n N o t e Brightfield and Fluorescence Imaging using 3D PrimeSurface Ultra-Low Attachment Microplates Brad Larson, BioTek Instruments, Inc., Winooski, VT USA Anju Dang, S-BIO, Hudson,

More information

Automated Image Processing to Quantify Cell Migration

Automated Image Processing to Quantify Cell Migration Automated Image Processing to Quantify Cell Migration Minmin Shen 1, Bastian Zimmer 2, Marcel Leist 2, Dorit Merhof 1 1 Interdiscipinary Center for Interative Data Analysis, Modelling and Visual Exploration

More information

Cellular Assays. A Strategic Market Analysis. Sample Slides

Cellular Assays. A Strategic Market Analysis. Sample Slides A Strategic Market Analysis Sample Slides 2002 For information contact: Frontline Strategic Consulting, Inc. 1065 E. Hillsdale Blvd, Suite 403, Foster City, CA 94404 650-525-1500 x125, x135 or x145 info@frontlinesmc.com

More information

Quality Control Assessment in Genotyping Console

Quality Control Assessment in Genotyping Console Quality Control Assessment in Genotyping Console Introduction Prior to the release of Genotyping Console (GTC) 2.1, quality control (QC) assessment of the SNP Array 6.0 assay was performed using the Dynamic

More information

Supporting Information

Supporting Information Supporting Information Jones et al. 10.1073/pnas.0808843106 SI Text Additional Data. Additional detailed supplemental data is available at www.cellprofiler.org/pnas2009.html. Features List shows the cytological

More information

Automated Image Analysis of Microstructure Changes in Metal Alloys

Automated Image Analysis of Microstructure Changes in Metal Alloys Automated Image Analysis of Microstructure Changes in Metal Alloys Mohammed E. Hoque, Ralph M. Ford, and John T. Roth Penn State Erie, The Behrend College School of Engineering and Engineering Technology

More information

Automated Imaging and Dual-Mask Analysis of γh2ax Foci to Determine DNA Damage on an Individual Cell Basis

Automated Imaging and Dual-Mask Analysis of γh2ax Foci to Determine DNA Damage on an Individual Cell Basis A p p l i c a t i o n N o t e Automated Imaging and Dual-Mask Analysis of γh2ax Foci to Determine DNA Damage on an Individual Cell Basis Brad Larson, BioTek Instruments, Inc., Winooski, VT USA Asha Sinha

More information

YIELD IMPROVEMENT CASE STUDY: STACKED SPRING CAPS

YIELD IMPROVEMENT CASE STUDY: STACKED SPRING CAPS YIELD IMPROVEMENT CASE STUDY: STACKED SPRING CAPS Shouzhu Ou 1, Kent Carlson 1, Malcolm Blair 2, Graham Jones 3, Richard Hardin 1 and Christoph Beckermann 4 1 Research Engineers, Department of Mechanical

More information

Why use Individual Bacteria Count instead of Colony Forming Units?

Why use Individual Bacteria Count instead of Colony Forming Units? Why use Individual Bacteria Count instead of Colony Forming Units? - Because it will assist the milk producer better in improving milk quality! In most countries worldwide the bacteriological quality of

More information

A Workflow to Characterize and Benchmark Human Induced Pluripotent Stem Cells Using the Operetta High-Content Screening System

A Workflow to Characterize and Benchmark Human Induced Pluripotent Stem Cells Using the Operetta High-Content Screening System CASE STUDY High-Content Analysis A Workflow to Characterize and Benchmark Human Induced Pluripotent Stem Cells Using the Operetta High-Content Screening System Human induced pluripotent stem cells (ipscs)

More information

Cover Page. The handle holds various files of this Leiden University dissertation.

Cover Page. The handle   holds various files of this Leiden University dissertation. Cover Page The handle http://hdl.handle.net/1887/22550 holds various files of this Leiden University dissertation. Author: Yan, Kuan Title: Image analysis and platform development for automated phenotyping

More information

4 Image Analysis of plastic deformation in the fracture of paper

4 Image Analysis of plastic deformation in the fracture of paper 4 Image Analysis of plastic deformation in the fracture of paper 4.1 Introduction As detailed in Chapter 2, one of the fundamental problems that arises in the estimation of the fracture toughness of an

More information

Automatic detection of plasmonic nanoparticles in tissue sections

Automatic detection of plasmonic nanoparticles in tissue sections Automatic detection of plasmonic nanoparticles in tissue sections Dor Shaviv and Orly Liba 1. Introduction Plasmonic nanoparticles, such as gold nanorods, are gaining popularity in both medical imaging

More information

Q: Using at least 3 biological replicates in an experiment is recommended to do. What do you suggest: At which step of calculation of the relative

Q: Using at least 3 biological replicates in an experiment is recommended to do. What do you suggest: At which step of calculation of the relative The questions below have been asked by attendees of the qpcr webinar series, Part 2: Analyzing Your Data. All the questions, including the questions that could not be answered during the webinar have been

More information

Single-Cell. Defy the Law of Averages

Single-Cell. Defy the Law of Averages Single-Cell AnalysiS Defy the Law of Averages An entirely new approach Single-Cell AnalysiS that Defies the Law of Averages Get Accurate, Reliable Gene expression Data with the fluidigm Single-Cell Workflow

More information

Data analysis in cell-based functional assay

Data analysis in cell-based functional assay 02.02.2005 Florian Hahne Molecular Genome Analysis Data analysis in cell-based functional assay Tools for automated pre-processing, analysis and visualization of high throughput FACS data Overview Challenge

More information

Chapter 3. Displaying and Summarizing Quantitative Data. 1 of 66 05/21/ :00 AM

Chapter 3. Displaying and Summarizing Quantitative Data.  1 of 66 05/21/ :00 AM Chapter 3 Displaying and Summarizing Quantitative Data D. Raffle 5/19/2015 1 of 66 05/21/2015 11:00 AM Intro In this chapter, we will discuss summarizing the distribution of numeric or quantitative variables.

More information

Gene Expression Data Analysis

Gene Expression Data Analysis Gene Expression Data Analysis Bing Zhang Department of Biomedical Informatics Vanderbilt University bing.zhang@vanderbilt.edu BMIF 310, Fall 2009 Gene expression technologies (summary) Hybridization-based

More information

Analysis of a Proposed Universal Fingerprint Microarray

Analysis of a Proposed Universal Fingerprint Microarray Analysis of a Proposed Universal Fingerprint Microarray Michael Doran, Raffaella Settimi, Daniela Raicu, Jacob Furst School of CTI, DePaul University, Chicago, IL Mathew Schipma, Darrell Chandler Bio-detection

More information

Automatic Epithelial Cells Detection of Pap smears images using Fuzzy C-Means Clustering

Automatic Epithelial Cells Detection of Pap smears images using Fuzzy C-Means Clustering 2012 4th International Conference on Bioinformatics and Biomedical Technology IPCBEE vol.29 (2012) (2012) IACSIT Press, Singapore Automatic Epithelial Cells Detection of Pap smears images using Fuzzy C-Means

More information

Feature Selection of Gene Expression Data for Cancer Classification: A Review

Feature Selection of Gene Expression Data for Cancer Classification: A Review Available online at www.sciencedirect.com ScienceDirect Procedia Computer Science 50 (2015 ) 52 57 2nd International Symposium on Big Data and Cloud Computing (ISBCC 15) Feature Selection of Gene Expression

More information

DEFY THE LAW OF AVERAGES. Single-Cell Targeted Gene Expression Analysis

DEFY THE LAW OF AVERAGES. Single-Cell Targeted Gene Expression Analysis DEFY THE LAW OF AVERAGES Single-Cell Targeted Gene Expression Analysis SINGLE-CELL ANALYSIS THAT DEFIES THE LAW OF AVERAGES GET ACCURATE, RELIABLE GENE EXPRESSION DATA WITH THE FLUIDIGM SINGLE-CELL WORKFLOW

More information

SURFACE ENHANCED RAMAN SCATTERING NANOPARTICLES AS AN ALTERNATIVE TO FLUORESCENT PROBES AN EVALUATION

SURFACE ENHANCED RAMAN SCATTERING NANOPARTICLES AS AN ALTERNATIVE TO FLUORESCENT PROBES AN EVALUATION APPLICATION NOTE SURFACE ENHANCED RAMAN SCATTERING NANOPARTICLES AS AN ALTERNATIVE TO FLUORESCENT PROBES AN EVALUATION Summary: Interest in using nanoparticles specifically, Surface Enhanced Raman Scattering

More information

Quantitative Real time PCR. Only for teaching purposes - not for reproduction or sale

Quantitative Real time PCR. Only for teaching purposes - not for reproduction or sale Quantitative Real time PCR PCR reaction conventional versus real time PCR real time PCR principles threshold cycle C T efficiency relative quantification reference genes primers detection chemistry GLP

More information

Background Analysis and Cross Hybridization. Application

Background Analysis and Cross Hybridization. Application Background Analysis and Cross Hybridization Application Pius Brzoska, Ph.D. Abstract Microarray technology provides a powerful tool with which to study the coordinate expression of thousands of genes in

More information

Image-based Quantification of Skin Irritation by Spatial Biomarker Profiling

Image-based Quantification of Skin Irritation by Spatial Biomarker Profiling Image-based Quantification of Skin Irritation by Spatial Biomarker Profiling Thora Pommerencke, Kathi Westphal, Claudia Ernst, Hartmut Dickhaus, Niels Grabe Institute for Medical Biometry and Informatics,

More information

Quality Measures for 24sure Microarrays

Quality Measures for 24sure Microarrays Quality Measures for 24sure Microarrays How to evaluate 24sure quality in BlueFuse Multi software. Introduction Data quality is one of the most important aspects of any microarray experiment. This technical

More information

Measuring gene expression

Measuring gene expression Measuring gene expression Grundlagen der Bioinformatik SS2018 https://www.youtube.com/watch?v=v8gh404a3gg Agenda Organization Gene expression Background Technologies FISH Nanostring Microarrays RNA-seq

More information

Strength in numbers? Modelling the impact of businesses on each other

Strength in numbers? Modelling the impact of businesses on each other Strength in numbers? Modelling the impact of businesses on each other Amir Abbas Sadeghian amirabs@stanford.edu Hakan Inan inanh@stanford.edu Andres Nötzli noetzli@stanford.edu. INTRODUCTION In many cities,

More information

Quality Measures for 24sure Microarrays

Quality Measures for 24sure Microarrays Quality Measures for 24sure Microarrays How to evaluate 24sure quality in BlueFuse Multi software. Introduction Data quality is one of the most important aspects of any microarray experiment. This technical

More information

Data Analysis on the ABI PRISM 7700 Sequence Detection System: Setting Baselines and Thresholds. Overview. Data Analysis Tutorial

Data Analysis on the ABI PRISM 7700 Sequence Detection System: Setting Baselines and Thresholds. Overview. Data Analysis Tutorial Data Analysis on the ABI PRISM 7700 Sequence Detection System: Setting Baselines and Thresholds Overview In order for accuracy and precision to be optimal, the assay must be properly evaluated and a few

More information

Super Resolution Imaging Solution Provider. Imaging Future

Super Resolution Imaging Solution Provider. Imaging Future Super Resolution Imaging Solution Provider Imaging Future Imaging Solution More Than Equipment NanoBioImaging(NBI) is the Industrial Partner of HKUST Super Resolution Imaging Center (SRIC). NBI aims to

More information

Multiplex Fluorescence Assays for Adherence Cells without Trypsinization

Multiplex Fluorescence Assays for Adherence Cells without Trypsinization Multiplex Fluorescence Assays for Adherence Cells without Trypsinization The combination of a bright field and three fluorescent channels allows the Celigo to perform many multiplexed assays. A gating

More information

NovoCyte Flow Cytometer

NovoCyte Flow Cytometer NovoCyte Flow Cytometer The Flow Cytometer for Everyone 2 Experience the NovoCyte Advantage Focus on advancing your research. Let the flow cytometer do the rest. NovoCyte Flow Cytometer High Performance

More information

Identification of biological themes in microarray data from a mouse heart development time series using GeneSifter

Identification of biological themes in microarray data from a mouse heart development time series using GeneSifter Identification of biological themes in microarray data from a mouse heart development time series using GeneSifter VizX Labs, LLC Seattle, WA 98119 Abstract Oligonucleotide microarrays were used to study

More information

QPix Systems. An industry standard for microbial colony picking. Genetix Now part of Molecular Devices.

QPix Systems. An industry standard for microbial colony picking. Genetix Now part of Molecular Devices. QPix Systems An industry standard for microbial colony picking www.moleculardevices.com/genetix Genetix Now part of Molecular Devices Setting an industry standard Although an essential step in many different

More information

Data Mining for Biological Data Analysis

Data Mining for Biological Data Analysis Data Mining for Biological Data Analysis Data Mining and Text Mining (UIC 583 @ Politecnico di Milano) References Data Mining Course by Gregory-Platesky Shapiro available at www.kdnuggets.com Jiawei Han

More information

OriGene GFC-Arrays for High-throughput Overexpression Screening of Human Gene Phenotypes

OriGene GFC-Arrays for High-throughput Overexpression Screening of Human Gene Phenotypes OriGene GFC-Arrays for High-throughput Overexpression Screening of Human Gene Phenotypes High-throughput Gene Function Validation Tool Introduction sirna screening libraries enable scientists to identify

More information

Enhancers mutations that make the original mutant phenotype more extreme. Suppressors mutations that make the original mutant phenotype less extreme

Enhancers mutations that make the original mutant phenotype more extreme. Suppressors mutations that make the original mutant phenotype less extreme Interactomics and Proteomics 1. Interactomics The field of interactomics is concerned with interactions between genes or proteins. They can be genetic interactions, in which two genes are involved in the

More information

Microarrays & Gene Expression Analysis

Microarrays & Gene Expression Analysis Microarrays & Gene Expression Analysis Contents DNA microarray technique Why measure gene expression Clustering algorithms Relation to Cancer SAGE SBH Sequencing By Hybridization DNA Microarrays 1. Developed

More information

Defense AT&L: May-June

Defense AT&L: May-June 28 Sizing the Supply Chain FROM THE TOP DOWN Rob Blakey Michael A. Bayer, Ph.D. Spare-part inventory has been, and continues to be, an important element of logistics support to the U.S. Air Force (USAF)

More information

Performance of cell viability and cytotoxicity assays on the IN Cell Analyzer 3000

Performance of cell viability and cytotoxicity assays on the IN Cell Analyzer 3000 GE Healthcare Application Note 28-4070-51 AA IN Cell Analyzer 3000 Performance of cell viability and cytotoxicity assays on the IN Cell Analyzer 3000 Key words: cell-based assay viability cytotoxicity

More information

Comp/Phys/Mtsc 715. Example Videos. Administrative 4/12/2012. Bioinformatics Visualization. Vis 2005, Bertram. Vis 2005, Cantarel(tighten.

Comp/Phys/Mtsc 715. Example Videos. Administrative 4/12/2012. Bioinformatics Visualization. Vis 2005, Bertram. Vis 2005, Cantarel(tighten. Comp/Phys/Mtsc 715 Bioinformatics Visualization Example Videos Vis 2005, Bertram Visualizing sound wavefront propagation Vis 2005, Cantarel(tighten.mov) Visualizing self contact in tightening knots Administrative

More information

COS 597c: Topics in Computational Molecular Biology. DNA arrays. Background

COS 597c: Topics in Computational Molecular Biology. DNA arrays. Background COS 597c: Topics in Computational Molecular Biology Lecture 19a: December 1, 1999 Lecturer: Robert Phillips Scribe: Robert Osada DNA arrays Before exploring the details of DNA chips, let s take a step

More information

The first thing you will see is the opening page. SeqMonk scans your copy and make sure everything is in order, indicated by the green check marks.

The first thing you will see is the opening page. SeqMonk scans your copy and make sure everything is in order, indicated by the green check marks. Open Seqmonk Launch SeqMonk The first thing you will see is the opening page. SeqMonk scans your copy and make sure everything is in order, indicated by the green check marks. SeqMonk Analysis Page 1 Create

More information

Mate-pair library data improves genome assembly

Mate-pair library data improves genome assembly De Novo Sequencing on the Ion Torrent PGM APPLICATION NOTE Mate-pair library data improves genome assembly Highly accurate PGM data allows for de Novo Sequencing and Assembly For a draft assembly, generate

More information

BIO4342 Lab Exercise: Detecting and Interpreting Genetic Homology

BIO4342 Lab Exercise: Detecting and Interpreting Genetic Homology BIO4342 Lab Exercise: Detecting and Interpreting Genetic Homology Jeremy Buhler March 15, 2004 In this lab, we ll annotate an interesting piece of the D. melanogaster genome. Along the way, you ll get

More information

Selected Techniques Part I

Selected Techniques Part I 1 Selected Techniques Part I Gel Electrophoresis Can be both qualitative and quantitative Qualitative About what size is the fragment? How many fragments are present? Is there in insert or not? Quantitative

More information

DNA Microarray Technology

DNA Microarray Technology 2 DNA Microarray Technology 2.1 Overview DNA microarrays are assays for quantifying the types and amounts of mrna transcripts present in a collection of cells. The number of mrna molecules derived from

More information

BIOINFORMATICS THE MACHINE LEARNING APPROACH

BIOINFORMATICS THE MACHINE LEARNING APPROACH 88 Proceedings of the 4 th International Conference on Informatics and Information Technology BIOINFORMATICS THE MACHINE LEARNING APPROACH A. Madevska-Bogdanova Inst, Informatics, Fac. Natural Sc. and

More information

Runs of Homozygosity Analysis Tutorial

Runs of Homozygosity Analysis Tutorial Runs of Homozygosity Analysis Tutorial Release 8.7.0 Golden Helix, Inc. March 22, 2017 Contents 1. Overview of the Project 2 2. Identify Runs of Homozygosity 6 Illustrative Example...............................................

More information

JCB. Supplemental material THE JOURNAL OF CELL BIOLOGY. Prospéri et al.,

JCB. Supplemental material THE JOURNAL OF CELL BIOLOGY. Prospéri et al., Supplemental material JCB Prospéri et al., http://www.jcb.org/cgi/content/full/jcb.201501018/dc1 THE JOURNAL OF CELL BIOLOGY Figure S1. Myo1b Tail interacts with YFP-EphB2 coated beads and genistein inhibits

More information

NEW INSIGHTS. NEW DISCOVERIES. Real-time automated measurements of cell health, movement and function inside your incubator.

NEW INSIGHTS. NEW DISCOVERIES. Real-time automated measurements of cell health, movement and function inside your incubator. THE NEXT GENERATION HAS ARRIVED IncuCyte S3 Live-Cell Analysis System Real-time automated measurements of cell health, movement and function inside your incubator. NEW INSIGHTS. NEW DISCOVERIES. See what

More information

The first and only fully-integrated microarray instrument for hands-free array processing

The first and only fully-integrated microarray instrument for hands-free array processing The first and only fully-integrated microarray instrument for hands-free array processing GeneTitan Instrument Transform your lab with a GeneTitan Instrument and experience the unparalleled power of streamlining

More information

Top-down Forecasting Using a CRM Database Gino Rooney Tom Bauer

Top-down Forecasting Using a CRM Database Gino Rooney Tom Bauer Top-down Forecasting Using a CRM Database Gino Rooney Tom Bauer Abstract More often than not sales forecasting in modern companies is poorly implemented despite the wealth of data that is readily available

More information

% Viability. isw2 ino isw2 ino isw2 ino isw2 ino mM HU 4-NQO CPT

% Viability. isw2 ino isw2 ino isw2 ino isw2 ino mM HU 4-NQO CPT a Drug concentration b 1.3% MMS nhp1 nhp1 8 nhp1 mag1.5% MMS.3% MMS nhp1 nhp1 ino8 9 ino8 9 % Viability 4.5% MMS ino8 9 ino8 9 2.5.1.15 % MMS c d nhp1 nhp1 nhp1 nhp1 nhp1 nhp1 Control (YPD) γ IR (1 gy)

More information

Automated Method for Determination of Infectious Dose (TCID 50 ) using Celigo Imaging Cytometer

Automated Method for Determination of Infectious Dose (TCID 50 ) using Celigo Imaging Cytometer Automated Method for Determination of Infectious Dose (TCID 50 ) using Celigo Imaging Cytometer Nexcelom Bioscience LLC. 360 Merrimack Street, Building 9 Lawrence, MA 01843 T: 978.327.5340 F: 978.327.5341

More information

Discovering gene regulatory control using ChIP-chip and ChIP-seq. Part 1. An introduction to gene regulatory control, concepts and methodologies

Discovering gene regulatory control using ChIP-chip and ChIP-seq. Part 1. An introduction to gene regulatory control, concepts and methodologies Discovering gene regulatory control using ChIP-chip and ChIP-seq Part 1 An introduction to gene regulatory control, concepts and methodologies Ian Simpson ian.simpson@.ed.ac.uk http://bit.ly/bio2links

More information

High-throughput physical phenotyping of cell differentiation

High-throughput physical phenotyping of cell differentiation Supplementary file High-throughput physical phenotyping of cell differentiation Jonathan Lin1,*, Donghyuk Kim1,*, Henry Tse2, Peter Tseng3, Lillian Peng1, Manjima Dhar1, Saravanan Karumbayaram4 and Dino

More information

Machine learning applications in genomics: practical issues & challenges. Yuzhen Ye School of Informatics and Computing, Indiana University

Machine learning applications in genomics: practical issues & challenges. Yuzhen Ye School of Informatics and Computing, Indiana University Machine learning applications in genomics: practical issues & challenges Yuzhen Ye School of Informatics and Computing, Indiana University Reference Machine learning applications in genetics and genomics

More information

How To Choose a GeneCopoeia Luciferase System. Ed Davis, Ph.D.

How To Choose a GeneCopoeia Luciferase System. Ed Davis, Ph.D. TECHNICAL NOTE How To Choose a GeneCopoeia Luciferase System Ed Davis, Ph.D. Introduction Luciferase reporter systems are invaluable tools for several applications, including regulation of gene expression

More information

OTC PP. Measuring Oil in Water: A Sanity Check Lew Brown, Mason Ide, and Peter Wolfe, Fluid Imaging Technologies, Inc.

OTC PP. Measuring Oil in Water: A Sanity Check Lew Brown, Mason Ide, and Peter Wolfe, Fluid Imaging Technologies, Inc. OTC-20192-PP Measuring Oil in Water: A Sanity Check Lew Brown, Mason Ide, and Peter Wolfe, Fluid Imaging Technologies, Inc. Copyright 2009, Offshore Technology Conference This paper was prepared for presentation

More information

A Systematic Approach to Performance Evaluation

A Systematic Approach to Performance Evaluation A Systematic Approach to Performance evaluation is the process of determining how well an existing or future computer system meets a set of alternative performance objectives. Arbitrarily selecting performance

More information

DNA/RNA MICROARRAYS NOTE: USE THIS KIT WITHIN 6 MONTHS OF RECEIPT.

DNA/RNA MICROARRAYS NOTE: USE THIS KIT WITHIN 6 MONTHS OF RECEIPT. DNA/RNA MICROARRAYS This protocol is based on the EDVOTEK protocol DNA/RNA Microarrays. 10 groups of students NOTE: USE THIS KIT WITHIN 6 MONTHS OF RECEIPT. 1. EXPERIMENT OBJECTIVE The objective of this

More information

Implementation of Qualitative Uncertainty Guidance: A Worked Example

Implementation of Qualitative Uncertainty Guidance: A Worked Example Implementation of Qualitative Uncertainty Guidance: A Worked Example Ruth Salway, Gavin Shaddick University of Bath Introduction The document 'Guidance on Qualitative Uncertainty Assessment' proposed a

More information

/ Computational Genomics. Time series analysis

/ Computational Genomics. Time series analysis 10-810 /02-710 Computational Genomics Time series analysis Expression Experiments Static: Snapshot of the activity in the cell Time series: Multiple arrays at various temporal intervals Time Series Examples:

More information

CELL CYCLE BASICS. G0/1 = 1X S Phase G2/M = 2X DYE FLUORESCENCE

CELL CYCLE BASICS. G0/1 = 1X S Phase G2/M = 2X DYE FLUORESCENCE CELL CYCLE BASICS Analysis of a population of cells replication state can be achieved by fluorescence labeling of the nuclei of cells in suspension and then analyzing the fluorescence properties of each

More information

Microarray. Key components Array Probes Detection system. Normalisation. Data-analysis - ratio generation

Microarray. Key components Array Probes Detection system. Normalisation. Data-analysis - ratio generation Microarray Key components Array Probes Detection system Normalisation Data-analysis - ratio generation MICROARRAY Measures Gene Expression Global - Genome wide scale Why Measure Gene Expression? What information

More information

Human SNP haplotypes. Statistics 246, Spring 2002 Week 15, Lecture 1

Human SNP haplotypes. Statistics 246, Spring 2002 Week 15, Lecture 1 Human SNP haplotypes Statistics 246, Spring 2002 Week 15, Lecture 1 Human single nucleotide polymorphisms The majority of human sequence variation is due to substitutions that have occurred once in the

More information

CAP BIOINFORMATICS Su-Shing Chen CISE. 10/5/2005 Su-Shing Chen, CISE 1

CAP BIOINFORMATICS Su-Shing Chen CISE. 10/5/2005 Su-Shing Chen, CISE 1 CAP 5510-9 BIOINFORMATICS Su-Shing Chen CISE 10/5/2005 Su-Shing Chen, CISE 1 Basic BioTech Processes Hybridization PCR Southern blotting (spot or stain) 10/5/2005 Su-Shing Chen, CISE 2 10/5/2005 Su-Shing

More information

Case Study: Dr. Jonny Wray, Head of Discovery Informatics at e-therapeutics PLC

Case Study: Dr. Jonny Wray, Head of Discovery Informatics at e-therapeutics PLC Reaxys DRUG DISCOVERY & DEVELOPMENT Case Study: Dr. Jonny Wray, Head of Discovery Informatics at e-therapeutics PLC Clean compound and bioactivity data are essential to successful modeling of the impact

More information

Survival Outcome Prediction for Cancer Patients based on Gene Interaction Network Analysis and Expression Profile Classification

Survival Outcome Prediction for Cancer Patients based on Gene Interaction Network Analysis and Expression Profile Classification Survival Outcome Prediction for Cancer Patients based on Gene Interaction Network Analysis and Expression Profile Classification Final Project Report Alexander Herrmann Advised by Dr. Andrew Gentles December

More information

Improving the Accuracy of Base Calls and Error Predictions for GS 20 DNA Sequence Data

Improving the Accuracy of Base Calls and Error Predictions for GS 20 DNA Sequence Data Improving the Accuracy of Base Calls and Error Predictions for GS 20 DNA Sequence Data Justin S. Hogg Department of Computational Biology University of Pittsburgh Pittsburgh, PA 15213 jsh32@pitt.edu Abstract

More information

AUTOMATIC DETECTION AND COUNTING OF PLATELETS IN MICROSCOPIC IMAGE 1. INTRODUCTION

AUTOMATIC DETECTION AND COUNTING OF PLATELETS IN MICROSCOPIC IMAGE 1. INTRODUCTION JOURNAL OF MEDICAL INFORMATICS & TECHNOLOGIES Vol. 16/2010, ISSN 1642-6037 pattern recognition, bioinformatic, machine learning, image analysis, platelet Robert BURDUK 1, Bartosz KRAWCZYK 1 AUTOMATIC DETECTION

More information

Expression summarization

Expression summarization Expression Quantification: Affy Affymetrix Genechip is an oligonucleotide array consisting of a several perfect match (PM) and their corresponding mismatch (MM) probes that interrogate for a single gene.

More information

PCR SYSTEMS. a new era in high-productivity qpcr. Applied Biosystems ViiA 7 Real-Time PCR System

PCR SYSTEMS. a new era in high-productivity qpcr. Applied Biosystems ViiA 7 Real-Time PCR System PCR SYSTEMS a new era in high-productivity qpcr Applied Biosystems ViiA 7 Real-Time PCR System a new era in high-productivity qpcr The ViiA 7 Real-Time PCR System delivers the proven reliability, sensitivity,

More information

Predicting Corporate Influence Cascades In Health Care Communities

Predicting Corporate Influence Cascades In Health Care Communities Predicting Corporate Influence Cascades In Health Care Communities Shouzhong Shi, Chaudary Zeeshan Arif, Sarah Tran December 11, 2015 Part A Introduction The standard model of drug prescription choice

More information

20.320, notes for 9/13

20.320, notes for 9/13 20.320 Notes Page 1 20.320, notes for 9/13 Thursday, September 13, 2012 9:36 AM Coming Assignments 1st part of design project coming up, assigned 9/14 (tomorrow) and due 9/21. Last time We covered ITC

More information

Exploration and Analysis of DNA Microarray Data

Exploration and Analysis of DNA Microarray Data Exploration and Analysis of DNA Microarray Data Dhammika Amaratunga Senior Research Fellow in Nonclinical Biostatistics Johnson & Johnson Pharmaceutical Research & Development Javier Cabrera Associate

More information

Toward Sharper Saws, Straighter Cuts, Higher Recovery: Managing the Process and Measuring the Results

Toward Sharper Saws, Straighter Cuts, Higher Recovery: Managing the Process and Measuring the Results Toward Sharper Saws, Straighter Cuts, Higher Recovery: Managing the Process and Measuring the Results Warren M. Bird California Saw & Knife Works San Francisco, California October, 2003 www.calsaw.com

More information

Using Mapmaker/QTL for QTL mapping

Using Mapmaker/QTL for QTL mapping Using Mapmaker/QTL for QTL mapping M. Maheswaran Tamil Nadu Agriculture University, Coimbatore Mapmaker/QTL overview A number of methods have been developed to map genes controlling quantitatively measured

More information

Chromosome-scale scaffolding of de novo genome assemblies based on chromatin interactions. Supplementary Material

Chromosome-scale scaffolding of de novo genome assemblies based on chromatin interactions. Supplementary Material Chromosome-scale scaffolding of de novo genome assemblies based on chromatin interactions Joshua N. Burton 1, Andrew Adey 1, Rupali P. Patwardhan 1, Ruolan Qiu 1, Jacob O. Kitzman 1, Jay Shendure 1 1 Department

More information

Normalization of Agilent Seahorse XF Data by In-situ Cell Counting Using a BioTek Cytation 5

Normalization of Agilent Seahorse XF Data by In-situ Cell Counting Using a BioTek Cytation 5 Normalization of Agilent Seahorse XF Data by In-situ Cell Counting Using a BioTek Cytation Application Note Authors Yoonseok Kam 1, Ned Jastromb 1, Joe Clayton, Paul Held, and Brian P. Dranka 1 1 Agilent

More information

Utility of Variable Bandwidth Monochromators for Quantification of Fluorescent Probes in Produced Effluent Water

Utility of Variable Bandwidth Monochromators for Quantification of Fluorescent Probes in Produced Effluent Water A p p l i c a t i o n N o t e Utility of Variable Bandwidth Monochromators for Quantification of Fluorescent Probes in Produced Effluent Water Brad Larson and Peter Banks, BioTek Instruments, Inc., Winooski,

More information

BIO 315 Lab Exam I. Section #: Name:

BIO 315 Lab Exam I. Section #: Name: Section #: Name: Also provide this information on the computer grid sheet given to you. (Section # in special code box) BIO 315 Lab Exam I 1. In labeling the parts of a standard compound light microscope

More information

Use of Phase Contrast Imaging to Track Morphological Cellular Changes due to Apoptotic Activity

Use of Phase Contrast Imaging to Track Morphological Cellular Changes due to Apoptotic Activity A p p l i c a t i o n N o t e Use of Phase Contrast Imaging to Track Morphological Cellular Changes due to Apoptotic Activity Brad Larson and Peter Banks, Applications Department, BioTek Instruments, Inc.,

More information

ab Propidium Iodide Flow Cytometry Kit for Cell Cycle Analysis

ab Propidium Iodide Flow Cytometry Kit for Cell Cycle Analysis ab139418 Propidium Iodide Flow Cytometry Kit for Cell Cycle Analysis Instructions for Use To determine cell cycle status in tissue culture cell lines by measuring DNA content using a flow cytometer. This

More information

MiniTEM. Designed for nanoparticle characterization

MiniTEM. Designed for nanoparticle characterization MiniTEM Designed for nanoparticle characterization MiniTEM revolutionizes access to transmission Transmission electron microscopy (TEM) is unmatched in providing high resolution images that allow visual

More information

ab Propidium Iodide Flow Cytometry Kit for Cell Cycle Analysis

ab Propidium Iodide Flow Cytometry Kit for Cell Cycle Analysis ab139418 Propidium Iodide Flow Cytometry Kit for Cell Cycle Analysis Instructions for Use To determine cell cycle status in tissue culture cell lines by measuring DNA content using a flow cytometer. This

More information

Automated Image Processing to Quantify Cell Migration

Automated Image Processing to Quantify Cell Migration Automated Image Processing to Quantify Cell Migration Minmin Shen!, Bastian Zimmer 2, Marcel Leist 2, Dorit Merhof 1 1 Interdiscipinary Center for Interative Data Analysis, Modelling and Visual Exploration

More information

If Dna Has The Instructions For Building Proteins Why Is Mrna Needed

If Dna Has The Instructions For Building Proteins Why Is Mrna Needed If Dna Has The Instructions For Building Proteins Why Is Mrna Needed if a strand of DNA has the sequence CGGTATATC, then the complementary each strand of DNA contains the info needed to produce the complementary

More information

TT SCORE. Trade Surveillance with Machine Learning THE NEED FOR TRADE SURVEILLANCE TRADITIONAL PARAMETER-BASED SURVEILLANCE TOOLS

TT SCORE. Trade Surveillance with Machine Learning THE NEED FOR TRADE SURVEILLANCE TRADITIONAL PARAMETER-BASED SURVEILLANCE TOOLS TT SCORE Trade Surveillance with Machine Learning THE NEED FOR TRADE SURVEILLANCE The financial industry has experienced great change and extraordinary challenges in the wake of the 2008 global financial

More information