search

austintwang / finemo_gpu

MIT License
5 stars 0 forks source link

No seqlets found in some loci within peaks #20

Closed kairukuma closed 4 months ago

kairukuma commented 4 months ago

Hello, this is great work and has helped me save a lot of time in identifying seqlet locations in my chrombpnet inputs. For the most part, finemo has been extremely easy to use, but I'm noticing that I'm missing outputs in some peaks. For example, I'm using the following interval as part of my input into chromBPnet:

chr18 61105509 61105961 mouse_Kupffer_ATAC_peak_103769 228 . 6.06604 22.8862 20.0038 358

As seen in the genome browser track linked below, the ATAC bigWig tracks associated with this interval show a small peak in the beginning followed by a larger peak:

https://genome.ucsc.edu/s/kylexwustl/mm10_chr18_issue

However, after running the call-hits method, the seqlet hits I get in this interval seem to only be in the larger peak:

chr18   61105763    61105773    pos_patterns.pattern_3  659.78217   -
chr18   61105778    61105788    pos_patterns.pattern_0  907.09644   -
chr18   61105817    61105827    pos_patterns.pattern_0  898.0232    +

Ideally, I'd like to find hits in both peaks without having to splice the original bed file and re-run chrombpnet contribs_bw + modisco motifs since these will require another day of CPU + GPU time. Is there a hyperparameter I can set to increase the sensitivity of these hits? Thanks a lot, any help is greatly appreciated!

kairukuma commented 4 months ago

A bit silly on my part by not reading the documentation fully. I ended up tweaking the alpha param down to 0.3 and got the desired output.