Mark A. DePristo

This paper uses deep learning to improve protein function annotation by training on Pfam data, achieving better accuracy and efficiency than BLASTp.

This paper develops a long-read sequencing protocol to improve variant detection and genome assembly in a human genome, achieving high accuracy and contiguity.

A convolutional network trained on 16S reads from 13,838 species reaches within 2.0% of training-set Bayes-optimal species accuracy and usually beats k-mer, alignment and Kraken2 on held-out-species genera.

The paper introduces CrowdVariant, a crowdsourcing framework to classify copy number variants (CNVs) using NGS data, and identifies 1,781 high-confidence CNVs in a reference sample.

A convolutional network on read-pileup images outperforms hand-built statistical variant callers, won an FDA challenge award for SNPs, learns new sequencing technologies and generalizes to other mammals.

Per-sample assembly with gVCF output, followed by joint genotyping, lets GATK HaplotypeCaller scale linearly with cohort size while calling indels more accurately than SAMtools, UnifiedGenotyper or Platypus.