Test filters on hybrid model

[iamciera]

Goall: Find out if the filters in the hybrid model are able to consistently result in similar sequences.

About: To understand what a filter is identifying a motif, you must input a sequence, and the filter identifies the main activation position, then you get the surrounding base pairs (bp) around that position on n length. In the example presented in lab meeting, @zhanyuanucb was using length 15bp.

How: Test filters similarity between different input sequences. The idea is that if you get the output of each filter from different input sequences, they should be the The different input sequences could be many things. @zhanyuanucb at first was using validation sequences from the original data that were ~1,000 bp.

Some more options that were discussed :

1. Make sure the input sequences are big so that you have good enough representation between sequences.

[zhanyuanucb]

1. What's been done by Apr 19:
   -inspect motifs from k folds of data

-generality of the model

2. What's still on progress by Apr 19:
   check the similarity of sequences

[iamciera]

I was thinking about the what the layers would be telling us and the "similarity" between sequences. What if the layers are just focusing on conserved regions between the sequences? How could we test this?