BPNet manuscript data 1

This repository holds the trained BPNet models and all the output files produced using the models including TF-MoDISco files and motif instances.

Files:

• output.tar.gz
  • Full output directory containing trained BPNet models, contribution scores, TF-MoDISco results, motif instances, etc.
  • See https://github.com/kundajelab/bpnet-manuscript for more information (archived at https://zenodo.org/record/4294814 with DOI: https://doi.org/10.5281/zenodo.4294813).
• bpnet.model.h5
  • BPNet Keras model. This model can be directly used via the model repository Kipoi: http://kipoi.org/models/BPNet-OSKN . 
• regions.bed
  • ChIP-nexus peak regions (resized to 1 kb width) where BPNet was trained and interpreted on.
  • Third column denotes the TF for which the peak was called.
• motif-instances.bed
  • BPNet motif instances for 11 representative motifs
  • Stored as BED file with the following columns
    • Chromosome
    • Start
    • End
    • Motif name
    • 'match' score - similarity to the CWM computed using the continuous Jaccard distance metric between the CWM and L1 normalized contribution scores
    • Strand
    • 'contrib' score - computed as the L1 norm of the contribution scores at the motif instance position
    • The log odds score with respect to the PWM derived from the PFM (i.e. classical PWM score)
• <TF>.preds.<strand>.bw
  • Model predictions in `regions.bed` for a particular TF in {Oct4, Sox2, Nanog, Klf4} and strand in {pos, neg}.
• <TF>.importance.counts.bw
  • Total count contribution scores for a particular TF
• <TF>.importance.profile.bw
  • Profile contribution scores. These were used to run TF-MoDISco and downstream analysis.
• PWM,CWM,ChIP-nexus-profiles.tar.gz
  • Directory containing PFM, CWM and aggregated ChIP-nexus footprint for 11 representative motifs.
• Figure-5d-periodicity.csv
  • Raw data corresponding to Figure 5d where the 10bp periodicity of contribution scores was computed for each motif.
• dfabf.Oct4-Sox2-subset.parq
  • Parquet file (read with `pd.read_parquet`) of genomics motif instance pairs and along with the BPNet-predicted interaction score.
  • See 08-motif-interactions.genomic.ipynb for more information on how to generate plots for Figure 5 from this data.