search

MiraldiLab / maxATAC

Transcription Factor Binding Prediction from ATAC-seq and scATAC-seq with Deep Neural Networks
Apache License 2.0
25 stars 8 forks source link

Reverse_complement training flag #45

Closed FaizRizvi closed 3 years ago

FaizRizvi commented 3 years ago

Currently rev_comp is hard coded into the code base we need to make it as a flag option similar to how shuffle is done.

Note*** when using rev_comp the total number epochs used for training should be doubled. keep this in mind while submitting bat files as well

This enhancement is needed to make final figures for publication

emiraldi commented 3 years ago

Hi Faiz! I don't understand why total number of epochs needs to be doubled. It made me worried that a single batch included both the forward and reverse strand for the same set of bins (as opposed to a random sample of bins with a mix of forward and reverse strands/directions). Can you fill me in Friday?

FaizRizvi commented 3 years ago

maxatac needs to see more examples of the reverse complement strand so that it can learn TF binding.

Wtih 50 epochs: 25 epochs are trained on the normal strand 25 epochs are trained on the reverse complement strand

Using 100 epochs doubles the number of epochs and more training examples are seen by maxatac.