CTCFMutants

This repository contains code for generating the results in Kaplow et al., "Neural network modeling of differential binding between wild-type and mutant CTCF reveals putative binding preferences for zinc fingers 1-2," BMC Genomics, 2022, including the wrapper for TF-MoDISco (Shrikumar et al., "TF-MoDISco v0.4.4.2-alpha: Technical Note," arXiv, 2018) that was used for obtaining TF-MoDISco motifs from deep convolutional neural networks trained to predict whether a CTCF ChIP-seq peak would have significantly lower in a dataset from CTCF with a mutated zinc finger as well as ipython notebooks for visualizing the TF-MoDISco results from those neural networks. It also contains scripts for analyses involving the TF-MoDISco results.

Code for General Use:

• runNewTFModisco.py: TF-MoDISco wrapper
• sequenceOperationsModiscoPrep.py: utilities used by runNewTFModisco.py
• ipython notebooks: code for visualizing results for each neural network, where the zinc finger number in the notebook name indicates the zinc finger mutant corresponding to the model; require data from http://mitra.stanford.edu/kundaje/imk1/CTCFMutantsProject/TFMoDIScoMotifs/
• convertFIMOToMotifHitBed.py: converts an output file from FIMO (Grant et al., "FIMO: Scanning for occurrences of a given motif," Bioinformatics, 2011) to a bed file ## Code for Analyses in Kaplow et al. (in evaluationScripts):
• analyzeCTCFsDataPlus.sh: code for analyses involving CTCF-s data (Le et al., "An alternative CTCF isoform antagonizes canonical CTCF occupancy and changes chromatin architecture to promote apoptosis," Nature Communications, 2019)
• analyzeCTCFUpstreamDownstreamNewTFModiscoMotifsAllHitsPlus.sh: code for analyses of mouse activated B cell peaks (Nakahashi et al., "A genome-wide map of CTCF multivalency redefines the CTCF code," Cell Reports, 2013) overlapping the core, upstream, and downstream motifs with no FIMO motif hit cutoff
• analyzeCTCFUpstreamDownstreamNewTFModiscoMotifsHeartAllHits.sh: code for analyses of mouse heart peaks (mouse ENCODE) overlapping the core, upstream, and downstream motifs with no FIMO motif hit cutoff
• analyzeCTCFUpstreamDownstreamNewTFModiscoMotifsHeartqVal.sh: code for analyses of mouse heart peaks overlapping the core, upstream, and downstream motifs with the FIMO motif hit q-value < 0.05 cutoff
• analyzeCTCFUpstreamDownstreamNewTFModiscoMotifsHeart.sh: code for analyses of mouse heart peaks overlapping the core, upstream, and downstream motifs with the default FIMO motif hit cutoff
• analyzeCTCFUpstreamDownstreamNewTFModiscoMotifsLiverAllHits.sh: code for analyses of mouse liver peaks (mouse ENCODE) overlapping the core, upstream, and downstream motifs with no FIMO motif hit cutoff
• analyzeCTCFUpstreamDownstreamNewTFModiscoMotifsLiverqVal.sh: code for analyses of mouse liver peaks overlapping the core, upstream, and downstream motifs with the FIMO motif hit q-value < 0.05 cutoff
• analyzeCTCFUpstreamDownstreamNewTFModiscoMotifsLiver.sh: code for analyses of mouse liver peaks overlapping the core, upstream, and downstream motifs with the default FIMO motif hit cutoff
• analyzeCTCFUpstreamDownstreamNewTFModiscoMotifsPlus.sh: code for analyses of mouse activated B cell peaks overlapping the core, upstream, and downstream motifs with the default FIMO motif hit cutoff
• analyzeCTCFUpstreamDownstreamNewTFModiscoMotifsqValPlus.sh: code for analyses of mouse activated B cell peaks overlapping the core, upstream, and downstream motifs with the FIMO motif hit q-value < 0.05 cutoff
• deseq2Script.r: code for obtaining differential peaks between wild type CTCF ChIP-seq and CTCF ChIP-seq with the zinc finger 1 mutant
• getDeepLiftScoresCTCFMutantsBigWigs.sh: code for obtaining DeepLIFT score bigwig files for each of the wild type CTCF verses mutant CTCF binding prediction models
• getDeepLiftScoresCTCFMutants.sh: code for obtaining DeepLIFT scores for each of the wild type CTCF versus mutant CTCF binding prediction models
• runNewTFModiscoCTCFMutants.sh: code for running TF-MoDISco on DeepLIFT scores for each of the wild type CTCF versus mutant CTCF binding prediction models ## Utilities for Code in evaluationScripts (in utils):
• getBestFIMOBed.py: gets the best motif hit from FIMO in a bed file
• makeViolinPlotsCoreDownstreamCTCFs.py: make violin plots for CTCFs analysis visualizations
• getDeepLIFTScores.py: wrapper for DeepLIFT (Shrikumar et al., "Learning important features through propagating activation differences," ICML, 2017) for models trained using Keras 0.3.2 with the Theano backend
• makeBedGraphFromPositionScores.py: makes a single bedGraph file from a text file with per-position DeepLIFT scores
• makeBedGraphFromPositionScoresPerSequence.py: makes a bedGraph file for each sequence from a text file with per-position DeepLIFT scores
• getDeepLIFTScoresCrossVal.py: wrapper for DeepLIFT for models trained using Keras 0.3.2 with the Theano background that iterates through cross-validation folds
• sequenceOperations.py: utilities for converting DNA sequence files into the numpy files for training deep learning models
• getDeepLIFTGrammars.py: wrapper for earlier version of TF-MoDISco that contains utilities for subsetting sequences based on deep learning model predictions ## Dependencies:
• python 2.7.15 (required for ipython notebooks, evaluationScripts, and utils) or 3.7.1
• numpy 1.14.3 (python 2) or 1.17.0 (python 3)
• matplotlib 2.2.3 (python 2) or 3.0.2 (python 3)
• h5py 2.6.0 (python 2) or 2.10.0 (python 3)
• seaborn 0.9.0 (required for only ipython notebooks)
• modisco 0.5.1.1 (python 2), 0.5.5.6 (python 2), or 0.5.14.1 (python 3)
• pybedtools 0.7.8 (python 2) or 0.8.1 (python 3)
• biopython 1.68 (python 2) or 1.73 (python 3)
• cython 0.29.12 (python 2) or 0.29.13 (python 3)
• meme 4.12.0 (evaluationScripts only)
• R 3.5.1 (evaluationScripts only)
• DESeq2 1.22.2 (evaluationScripts only)
• pybedtools 0.7.8 (evaluationScripts only)
• deeplift 0.5.5-theano (evaluationScripts only)
• keras 0.3.2 (evaluationScripts only)
• bedGraphToBigWig (http://hgdownload.soe.ucsc.edu/admin/exe/linux.x86_64/bedGraphToBigWig) (evaluationScripts only)
• Biopython 1.68 (evaluationScripts only)
• cython 0.29.12 (evaluationScripts only) ## Citation: Kaplow IM, Banerjee A, Foo CS. Neural network modeling of differential binding between wild-type and mutant CTCF reveals putative binding preferences for zinc fingers 1-2. BMC Genomics, 23: 295, 2022. ## Contact: Irene Kaplow: ikaplow@cs.stanford.edu

Chuan Sheng Foo: csfoo@cs.stanford.edu