Back

Learning What a Good Structural Variant Looks Like.

Chowdhury, M.; Layer, R. M.

2020-05-23 bioinformatics
10.1101/2020.05.22.111260 bioRxiv
Show abstract

Structural variations (SVs) are an important class of genetic mutations, yet SV detectors still suffer from high false-positive rates. In many cases, humans can quickly determine whether a putative SV is real by merely looking at a visualization of the SVs coverage profile. To that end, we developed Samplot-ML, a convolutional neural network (CNN) trained to genotype genomic deletions using Samplot visualizations that incorporate various forms of evidence such as genome coverage, discordant pairs, and split reads. Using Samplot-ML, we were able to reduce false positives by 47% while keeping 97% of true positives on average across several test samples.

Matching journals

The top 1 journal accounts for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.