This repository accompanies Gkmexplain: Fast and Accurate Interpretation of Nonlinear Gapped k-mer SVMs by Shrikumar*†, Prakash* and Kundaje† (*co-first authors †co-corresponding authors). Accepted to ISMB 2019 and published in Bioinformatics. DOI: 10.1093/bioinformatics/btz322
Video of ISMB 2019 talk: https://www.youtube.com/watch?v=bEDx_LjNk50
Please see lsgkmexplain_TALGATA.ipynb and lsgkmexplain_Nanog.ipynb for example notebooks demonstrating usage on simulated and real genomic data, respectively.
Please see https://github.com/kundajelab/gkmexplain/tree/master/dsQTL for scripts to replicate the dsQTL analysis.
Please see https://github.com/kundajelab/gkmexplain/tree/master/talgata for scripts to replicate the ananlysis on the simulated sequences containing TAL1 and GATA motifs.
Please see https://github.com/kundajelab/gkmexplain/tree/master/Nanog_H1ESC for information on how to replicate the ananlysis on the Nanog data. Note that the notebook lsgkmexplain_Nanog.ipynb contains the code to process the data, train the model, compute importance scores and run TF-MoDISco.
The code that implements the gkmexplain interpretation method (for genomic sequences) is at https://github.com/kundajelab/lsgkm (see gkmexplain.c under "src").
An example of using Integrated Gradients on a non-genomics dataset (the Madelon dataset) is at https://colab.research.google.com/drive/1LUGMIwOHLKDdK3deg7IIZ71kb2Nu1qv2