DNA Based Disease Prediction using pathway Analysis

Transcription

1 2017 IEEE 7th International Advance Computing Conference DNA Based Disease Prediction using pathway Analysis Syeeda Farah Dr.Asha T Cauvery B and Sushma M S Department of Computer Science and Shivanand K S Department of Computer Science and Engineering Department of Computer Science and Engineering Bangalore Institute of Technology and Engineering Bangalore Institute of Technology Bangalore, Karnataka Bangalore Institute of Technology Bangalore, Karnataka asha.masti@gmail.com Bangalore, Karnataka syeeda.farah.93@gmail.com cauverysoni@gmail.com sushmasuresh93@gmail.com k.s.shivanand23@gmail.com ABSTRACT Most diseases are not triggered by a single genome but by a combination of genomes together. Sequences occurring more frequently in the diseased samples than in the healthy samples indicate the generic factors of the disease. DNA has become an extremely useful tool for predicting disease. By allowing medical professionals to identify genes in DNA that are markers for diseases, a person can make appropriate lifestyle or similar modifications to help lower the risk of disease. We propose a system in which the above knowledge is provided by determining the probabilistic levels of a disease occurring if the causal gene or the associated genes are mutated. Index Terms Data Mining, Bayesian Network, Pathway Analysis, Disease Prevention. I. INTRODUCTION DNA is a molecule that encodes the genetic instructions used in the development and functioning of all known living organisms and viruses. DNA has become an extremely useful source for identifying and predicting diseases. The human genome includes approximately genes. With the exception of identical twins, no two humans have the same genome. The genetic information in a genome is held within genes, and the complete set of this information in an organism is called its genotype. A gene is a unit of heredity and is a region of DNA that influences a particular characteristic in an organism. A Single Nucleotide Polymorphism (SNP) is a DNA sequence variation occurring commonly with a population in which a single nucleotide-a, T, C or G-in the genome differs between members of a biological species or paired chromosomes. Where there is life there is always disorder or disease. Diseases can be broadly classified as those which are caused by viruses and bacteria, and those which are caused by malfunction of organs i.e., ailments. Mutation in DNA causes the diseases of the second type which remain unknown till the last stages where very little can be done to prevent them. The mutations in disease-causal-gene along with other supporting genes make the organ vulnerable hence leading to diseases. The presence of mutation in only the causal gene may or may not produce the diseases. The mutation in causal gene might have been induced due to mutation in the support gene. We consider all these factors. Our proposed system only takes into consideration those genes which can impact on the cause of the diseases to a fairly large extent and hence removes the negligible data, which is done by applying the concepts of Bayesian network specifically using the pathway analysis concepts which provides a pathway of associated causal and supporting genes. This Paper starts by describing the necessary information and background details that are needed for understanding the work done. Followed by the methodologies that have been used in the proposed work. We then move on to the architecture of the proposed work. In this section we explain the algorithm and the implementation of the concepts we learned in the previous section. Followed by the Results of the case studies on diseases like Type-1 Diabetes and Crohn s disease. After this we have the conclusion and references. II. BACKGROUND A prior biological knowledge was needed to understand the genetics behind various diseases. Genetic diseases in particular. We went about every chromosome and the disease causing genes on every chromosome. Further, we studied gene interaction patterns in many diseases such as Crohn s disease, Alzheimer s disease, and a few types of cancers, etc. We found [7] that a disease can be caused by a mutation in a single gene or by a collection of genes. We also found [3] that mutation of one gene can be responsible for two or more diseases (Like Alzheimer s and Parkinson s diseases). Another interesting finding we found is that a typical microarray [3] data may have only a small number of records, while the number of fields, corresponding to number of genes is in thousands. Having so many fields relative to few samples creates a high likelihood of finding false positives. We decided to use the concepts of Pathway analysis in data mining because [6] Pathway analysis has become the first choice for extracting and explaining the underlying biology for high throughput molecular measurements. Today, virtually every bioinformatics study looks for statistically significant pathways as either biological interpretation or validation of computationally derived results /17 $ IEEE DOI /IACC

2 Hence aiding us to reduce these false positives and obtain a higher accuracy result. III. MODULES components denote the joint probability distribution for X. The BN structure S is a directed acyclic graph, meaning that the network is hierarchical and has both top-level and terminal nodes and no directed paths which eventually return to them. Given structure S, the joint probability distribution for X is given by, p (x) = (2) Score-based methods consider a number of possible BN structures and assign a score to each that measures how well it explains the observed set of data. The algorithms then typically return the single structure that maximizes the score. AIC= log (p (D, G )) - (3) Fig. 1. Architecture diagram A. Data collection The genes are extracted from the NCBI Datasets in the GenBank format and required fields are extracted for the statistical calculations. The GenBank format of genetic data will include various details such as gene name, NCBI gene ID, location if that gene on the chromosome along with the base pair ordering, species, etc. Along with these details, it contains the respective gene sequence. We have also integrated other databases to obtain the correlation values of the diseasomes. B. Data-Preprocessing GenBank data was processed using Perl program to extract gene ID and gene name. Results were tabulated in a database table which will be used further to visualize pathways.correlation is the proportion of variance that two traits share due to genetic causes.from correlation, the p-values were calculated and fed into the Bayesian network to build pathways (using JDBC). The Correlation co-efficient for a pair of genes is calculated using the Pearson s correlation coefficient as: Basic outline of Bayesian network: Input: Observational data Output: Bayesian network Input a cut-off value to restrict the number of records/variables which is provided as input to the algorithm. Use p-values from the database table obtained as an output from data processing. Generate the initial BN, evaluate and set it as the current BN(Using an appropriate scoring function) Generate the initial BN, evaluate and set it as the current BN(Using an appropriate scoring function) If the score of the neighbor is better than the score of the current BN, set the neighbor with the best score as the current BN and return to step 3. Evaluate the neighbors of the current BN. Otherwise stop the learning process. Visualize the pathway using JGraph like package. D. Visualizing the Pathway Pathways are visualized using Java APIs like swings taking the network developed in the previous module as the input here. = (1) C. Building Bayesian Network Bayesian networks have a number of features that make them viable for combining prior knowledge and data as BNs can deal with uncertainty, avoid over-fitting a model to training data and learn from incomplete datasets. BNs handle stochastic events in a probabilistic framework. Many BN structure learning algorithms are based on heuristic search techniques with likelihood approximation because of the infeasible computational complexity. BNs are graphical representations of statistical interdependencies amongst sets of nodes. Specifically, a BN for a set of variables X = {X1, X2,..., Xn} consists of (1) a network structure S that encodes a set of conditional independence assertions about variables in X, and (2) a set P of conditional probability distributions associated with each variable (Heckerman, 2008). Together, these

3 IV. ARCHITECTURE Fig. 3. Detailed flow diagram for data pre-processing B. Build Bayesian Network. Fig. 2. Flow Diagram A. Data Preprocessing Gene Sequence in GenBank format is obtained from the NCBI website. The relevant gene ID and gene name is obtained for the particular gene sequence in question. The code for this is present in the Perl module. The perl module also fetches the correlation values for the associated genes of the causal gene. This is obtained from the co-expressed gene files named with gene IDs. The correlation values obtained are used to calculate the p values. This information is stored in the database in the form of tables. The first table created is the parent gene table. The tables are created recursively for every entry in the parent gene table. Fig. 4. Detailed flow diagram to build Bayesian network The Bayesian network is built using Java. This module uses the concepts of linked hash map. The java code developed has two methods calculateaic() and conditionalp(). calculateaic() takes in p values and calculates AIC(Akaike Information criterion). AIC is a scoring function used to determine the addition of genes into the network. public static int calculateaic(double p1) { double x; int k=n; x=(2*k-(2*math.log(p1))); int z=(int)x; return z; } conditional() takes in the p values of the parent node and child node. Here we calculate the probability of occurrence of the child node when the parent node has already occurred. This method returns the calculated p values. public static double conditionalp(double parentpval, double childpval) { Double condp=parentpval*childpval; return condp; } C. Visualizing the network. The linked hash map is accessed to visualize the pathway. This is done by using the JGraphx package in netbeans. This module is included in the Java code of the Bayesian network

4 V. CASE STUDY From the implementation of the Pathway analysis we have visualized the pathways between the genes and hence knowledge is provided about which genes can affect a disease collectively. This is depicted by taking Type 1 Diabetes into consideration. 1) Causal Gene: From the research we have found that HLA-DQB1 is the causal gene for Type 1 Diabetes. 2) Support Genes: Through data mining techniques we have obtained the support genes that cause the Type 1 Diabetes even if the Causal gene is not mutated. We use the Net Beans IDE to run all the modules mentioned in the previous section. The Perl program gives the desired records from the huge amount of data based on the user input (Gene name and number of records to be fetched). Further modules work on the tables fetched to give the output. The output is in the form of a network or a pathway of the causal gene and its associated gene. The causal gene is colored white and the genes associated closely with the causal gene are colored pink. The genes which are known to have an effect on the closely associated genes are colored cyan. We specify the causal gene name and the number of records to be evaluated as shown below in Fig. 5 Fig. 6. Disease pathway of HLA-DQB1 which causes Type 1 Diabetes (for 10 records) Fig. 5. Input screen and number of records to be evaluated Once the input is given, the program runs and the pathway is displayed. The pathway of the causal gene and its associated gene is shown in the Fig.6 for 10 records and Fig. 7 for 25 records respectively: Fig. 7. Disease pathway of HLA-DQB1 which causes Type 1 Diabetes (for 25 records) The following tables are created when we run the program: TABLE I. DATABASE TABLE CREATED FOR INPUT GENE I.E., HLA-DQB

5 TABLE II. Database table created for HLA-DQA1 TABLE V. Database table created for HLA-DRB6 TABLE III. Database table created for HLA-DRB1 Figure 8 shows the output for Sickle cell anemia which is caused by a single gene HBB. Since there are no associations in the onset of the disease only a single node is displayed. TABLE IV. Database table created for HLA-DPA1 Fig. 8. Disease pathway of HBB which causes Sickle cell anemia

6 VI. CONCLUSION A single gene may or may not cause a disease. Gene to gene interaction also should be considered because they can affect the functionality of the genes. To identify this phenomenon we have visualized pathways for each gene responsible for a disease and the probabilistic statistics of occurrence of a disease is predicted. The statistics have been obtained by eliminating the redundant data and taking into account only those genes in close correlation. REFERENCES [1] Li Ding, Michael C. Wendl, Daniel C. Koboldt and Elaine R. Mardis, Analysis of next-generation genomic data in cancer: accomplishments and challenges, Human Molecular Genetics, R1 R9, [2] Sebastian Okser, TapioPahikkala and TeroAittokallio, Genetic variants and their interactions in disease risk prediction machine learning and network perspectives, BioData Mining, 6:5, 1-16, [3] W. B. Langdon and B. F. Buxton, Genetic Programming for Mining DNA Chip Data from Cancer Patients, Genetic Programming and Evolvable Machines, 5, , [4] Davnah Urbach and Jason H Moore, Mining the diseasome, BioData Mining, 4:25, 1-2, [5] Vijay K Ramanan, Li Shen, Jason H. Moore and Andrew J. Saykin, Pathway analysis of genomic data: concepts, methods, and prospects for future development, National Institute of Health,28(7): , [6] Purvesh Khatri, MarinaSirota and Atul J. Butte, Ten Years of Pathway Analysis: Current Approaches and Outstanding Challenges, Computational Biology, 8:2, 1-11, [7] Qingrun Zhang, Quan Long1 and JurgOtt, AprioriGWAS, a New Pattern Mining Strategy for Detecting Genetic Variants Associated with Disease through Interaction Effects, Computational Biology, 10:6, 1-11, 2014 [8] Wei Zhang, Data mining for biological data learning algorithm and Application, A Dissertation Submitted to the Graduate School of the University of Notre Dame,