After a long hiatus, I’m going to write about another paper I read recently, that is changing the way I think about using deep networks for biological sequence problems.

Background

Anyone that’s working in applied machine learning these days is familiar with the idea of convolutional neural networks …

I've been reading a few papers recently that each involve training a deep network that takes input directly from sequence and predicts some aspect of chromatin. Deep Bind predicts protein-DNA binding, and DeepSEA predicts the effect of SNPs on chromatin state. The most recent paper, and the one I find …