GIA as a interpretation pipeline/package for pytorch?

p-koo / residualbind

"Global Importance Analysis: A Method to Quantify Importance of Genomic Features in Deep Neural Networks" by Koo et al.

MIT License

10 stars 3 forks source link

GIA as a interpretation pipeline/package for pytorch? #2

Open lejecrs opened 4 months ago

lejecrs commented 4 months ago

Hi! I really love the idea of GIA compared to attribution methods. Any consideration for compacting GIA as an interpretation pipeline/package for pytorch?

p-koo commented 4 months ago

Hi! Thanks for your interest. Yes, we can extend this to PyTorch. I’m thinking of redoing this as a framework agnostic tool. Due to other constraints, I’m hoping to have this by June. Will let you know if completed sooner. In the mean time, I’m happy to assist you if you have specific questions.

lejecrs commented 4 months ago

Thanks for the reply! I am interested in the motif-motif interaction in genomic models and want to first use the GIA as a motif finder and compare it with attribution-based methods. I also love your idea on GLIFAC (Ghotra et al.), any coding sample available for it?

p-koo commented 4 months ago

If interested in GLIFAC, please reach out to Chandana (rajesh@cshl.edu, cc koo@cshl.edu), she has taken over the project and extended it, including user-friendly code. We are working towards a full paper with more tricks involved, but would be happy to share preliminary code if you find it useful.