Prediction of transcription factor binding based on DNA sequence

Machine learning notebooks

• Classifying sequences as bound or unbound using only sequence data (target is either 0 or 1) Binary_Classification_without_shapes.ipynb
  • Same but with shapes Binary_WithShapes.ipynb
  • Same but with hot encoded target (not sure why I didn't do that in the other ones) Softmax_without_shapes.ipynb
• Regression analysis of actual binding sequences OnlyPositiveSamples.ipynb
  • Same but negative samples have a target of 0 PositiveAndNegative.ipynb
  • Same but with random sequences with a target of 0 PositiveAndRandom.ipynb

MATLAB for pre-processing the data

• main.mlx Get all the binding and non-binding sites for a particular TF, includes figures and instructions throughout
  • predict_positions.m Function that takes as input a full chromosome sequence, a PWM and a score threshhold and returns the positions of all the regions from that sequence that have a score above the threshhold.
    • get_score.m Function that returns the score for a possible binding sequence given a PWM.
  • check_found.m Filter the sequences into binding and not binding
• getTFsequences.mlx Get all the actual binding sequences for a particular transcription factor and analyze them.
• calculate_shapes.mlx Calulate DNA physical shape using pentamer method
  • create_pentamer_dict.mlx Create a dictionary of all possible pentamers for the shape calculations

Data Sources

• The sequence of the human genome (assembly hg19), available at: http://hgdownload.soe.ucsc.edu/goldenPath/hg19/bigZips/chromFa.tar.gz The format is self-explanatory. Lower-case letters correspond to repetitive regions, uppercase to non-repetitive regions. For this project, you should consider both as the same.
• A set of genomic regions that are active regulatory regions in a particular cell type (GM12878, which is a cell line derived from lymphoblasts). This is available here:
  • http://www.cs.mcgill.ca/~blanchem/561/wgEncodeRegTfbsClusteredV3.GM12878.merged.bed.gz
  • Stored here as wgEncodeRegTfbsClusteredV3.GM12878.merged.bed
• A set of genomic coordinates of actual transcription factor binding sites for several transcription factors:
  • http://www.cs.mcgill.ca/~blanchem/561/factorbookMotifPos.txt.gz
  • Stored here as factorbookMotifPos.txt
  • Each line looks like this:
    • 585 chr1 16245 16260 CTCF 1.97 -
    • Field 1: Ignore
    • Field 2,3,4: Genomic coordinate
    • Field 5: Name of transcription factor
    • Field 6: Score of predicted binding site (could probably be ignored)
    • Field 7: DNA strand (+ or -) of binding site
• Position weight matrices for each TF is available here:
  • http://www.cs.mcgill.ca/~blanchem/561/factorbookMotifPwm.txt.gz
  • Stored here as factorbookMotifPwm.txt
• Predicted structural properties for every human genome positions are available here: http://rohsdb.cmb.usc.edu/