kundajelab / bpnet

Toolkit to train base-resolution deep neural networks on functional genomics data and to interpret them
http://bit.ly/bpnet-colab
MIT License
141 stars 33 forks source link

contrib.counts vs contrib.profile #39

Closed ChangxuFan closed 2 years ago

ChangxuFan commented 2 years ago

Dear developers,

Thanks for developing the package! A quick question: bpnet export-bw generates 2 types of bw files, with suffixes contrib.counts and contrib.profile, respectively. I was wondering what the suffixes mean and what's the difference between them?

Thanks!!

Avsecz commented 2 years ago

Hey, we compute two differnet kinds of contribution scores in BPNet: those with respect to the profile prediction head (contrib.counts) and those w.r.t. total count predictions contrib.counts. Please see the bpnet paper for more information https://www.biorxiv.org/content/10.1101/737981v3.

ChangxuFan commented 2 years ago

The paper is very clear. Thanks!!