BST 226 Statistical Methods for Bioinformatics David M. Rocke. February 10, 2014 BST 226 Statistical Methods for Bioinformatics 1

Transcription

1 BST 226 Statistical Methods for Bioinformatics David M. Rocke February 10, 2014 BST 226 Statistical Methods for Bioinformatics 1

2 Proteomics by LS/MS/MS Proteins are digested with trypsin or other proteases so that the peptides are shorter. Trypsin cleaves peptide chains mainly at the carboxyl side of the amino acids lysine or arginine, except when either is followed by proline. Liquid Chromatography separates a mixture of peptides. The output of LC is periodically injected into the mass spec by electro-spray ionization, usually at millisecond intervals February 10, 2014 BST 226 Statistical Methods for Bioinformatics 2

3 Proteomics by LS/MS/MS In MS/MS (tandem mass spectrometry), the first stage identifies the mass to charge (m/z) of the components and injects the highest abundance components into a second stage. In the first stage, the instrument contains a hard vacuum. In the second stage, the compartment has a low amount of argon or xenon. Collisions with these neutral atoms cause fragmentation of the peptides. Noble gases such as argon and xenon are non-reactive chemically, so do not bind to the peptides. February 10, 2014 BST 226 Statistical Methods for Bioinformatics 3

4 February 10, 2014 BST 226 Statistical Methods for Bioinformatics 4

5 Fragmentation Patterns For a given peptide, the fragmentation patterns can be derived experimentally, but usually come from computation. The main matching is on the m/z of the fragments. A confidence value is assigned based on matching the pattern from the best peptide match vs. other possible peptides or from random assignment. For each identified peptide, it can be assigned uniquely to a protein or else to multiple proteins. February 10, 2014 BST 226 Statistical Methods for Bioinformatics 5

6 Spectral Counts The number of times a particular peptide is seen is allocated to the protein it matches, or else the count is split among proteins if it matches several. The spectral count for a given protein is the sum of the counts for all peptides matching to that protein. There is theory and experimental evidence that the spectral count is a reasonable relative estimate of the amount of that protein in the original sample. There is work ongoing towards using peak area in the first stage MS for quantitation, but most still use the spectral counts. February 10, 2014 BST 226 Statistical Methods for Bioinformatics 6

7 February 10, 2014 BST 226 Statistical Methods for Bioinformatics 7

8 Chicken Proteomics Data Chicken Corneocyte Cross-Linked Proteome, Robert H. Rice, Brett R. Winters, Blythe P. Durbin-Johnson, and David M. Rocke, J. Proteome Res., 2013, 12 (2), pp samples Four chickens Beak, Claw, Feather, Scale Soluble fraction, insoluble fraction 224 identified proteins February 10, 2014 BST 226 Statistical Methods for Bioinformatics 8

9 Chicken.Proteomics.xslx Column Content 1 (A) Row Number (Obs) 2 (B) Protein 3 (C) Accession Number 4 (D) MW in kda 5 36 (E AJ) Sample number 1 32 February 10, 2014 BST 226 Statistical Methods for Bioinformatics 9

10 Chicken.Factors.txt 32 by 3 table with a header row Columns are Chicken (1-4), Component (Beak, Claw, Feather, Scale), and Fraction (Soluble, Insoluble). February 10, 2014 BST 226 Statistical Methods for Bioinformatics 10

11 Data processing Read Chicken.Proteomics.xsls into R You could save the sheet as a tab delimited text file and read with read.delim() You could save the sheet as a.csv file and read with read.csv() You could install and load the CRAN library xlsx You need to install java for this to work chick <- read.xlsx( Chicken.Proteomics.xlsx", sheetindex = 1, stringsasfactors = F) Extract columns 5 36 as a matrix of counts February 10, 2014 BST 226 Statistical Methods for Bioinformatics 11

12 Data processing Read Chicken.Factors.txt using read.delim(). Replace the Chicken variable with a factor (otherwise Chicken is a number). Omit proteins (rows of the data matrix) where the number of zeroes is too high (for example, greater than 75% of the sample size, which is (0.75)(32) = 24. For each remaining protein, fit a glm model with the quasipoisson family and main effects only using the three variables. February 10, 2014 BST 226 Statistical Methods for Bioinformatics 12

13 Data processing For each remaining protein, fit a glm model with the quasipoisson family and main effects only using the three variables. Save the p-values for component and fraction. Use p.adjust() to get the FDR adjusted p-values, and select the proteins that have FDR adjusted p-values less than 0.10 for either component or fraction. Install the package multcomp, which allows the use of post-hoc tests for many different regression-type estimators. February 10, 2014 BST 226 Statistical Methods for Bioinformatics 13

14 Assignment for Wednesday Install xlsx and java if you wish. Read the data and the variables into R Remove the proteins with too many zero entries. Install multcomp Start thinking about how to code the glm loop. February 10, 2014 BST 226 Statistical Methods for Bioinformatics 14