- Fixed issue #150, regions_used not found while generating bigwigs from impotance h5s
- (MAJOR) Bug in chrombpnet modisco_motifs command. seqlets was limited to 50000. If users wanted to change it to 1 million this did not happen.
- Filter peaks at edges for pred_bw command and bias pipleline. So bias evaluation now done on these filtered peaks.
- Preprocessing deafulted to use unix sort. Provided option to switch to bedtools sort.
- Provided option to use filter chromosomes option in preprocessing.
- Changed pipelines to use modisco-lite, old modisco will soon be removed
- Added automatic shift scripts to repo and integrated with the pipeline
- modisco now outputs both html and pdf. pdf can be shared with anyone.
- Use ATAC, DNASE not ATAC_PE, DNASE_SE anymore.
- Simplyfing workflows to include only two main workflows, chrombpnet_train_tf_model, chrombpnet_train_bias_model
- Restructuring README, moving tutorial to additional documentation and introducing FAQ, and only two pipelines (chrombpnet_train_tf_model, chrombpnet_train_bias_model)
- Typo fix - (PR#31-36)
- We can now specify ylimit in marginal footprinting plots (PR#27)
- PR#30 merged to do fast genome wide gc binning and bug fix to ensure case-sensitve GC calculation. Adds a unit test to make checks.
- Updated modisco version - (PR#25)
- Typo fix - (PR#23-24)
- Marginal footprinting no longer hardcoded to take only chr1, it now inputs the fold json. (PR#20)
- Invivo footprinting is made compatible with the new input repo changes. (PR#22)
- Pandas read bugs fixed for gc-matching scripts. (PR#15)
- Tutorial now exits when one part of the pipeline breaks instead of proceeding further. (PR#18)
- Added the fast binning scripts at src/helpers/make_gc_matched_negatives/get_genomewide_gc_buckets (PR#11)
- Added pseudocount in metrics.py (PR#13) - needs further review.
- Seed setting fixed for training. Seed was set improperly - preventing shuffling for every epoch. (PR#19)
- return coordinates while training for debugging - returns status of the point such as revcomp,peak/nonpeak,coordinates etc
- In
step6_train_chrombpnet_model
changed the file saving name bug - Changed scripts in
src/evaluatuion/make_bigwigs/
to be compatible with new dir structure
- Added a note in README that users can continue to use the pre-trained bias model from chrombpnet-lite repo, if thats what they have been using till now
- Changed the default modisco crop setting in tutorial from 1000 to 500
- Fixed a bug in
src/evaluation/modisco/run_modisco.py
in directory creation