Search

GWAS often assume a linear phenotype-covariate relationship which may not hold in practice. Here the authors present DeepNull, in which they apply deep learning to identify and adjust for complex non-linear relationships, improving phenotypic prediction and GWAS power.

Deep learning reduces errors in sequences from PacBio HiFi reads.

A deep convolutional neural network calculates liability scores for chronic obstructive pulmonary disease (COPD) from raw spirogram traces and noisy medical-record-based labels in the UK Biobank. Genome-wide analyses using these scores replicate known loci for lung function and identify 67 new disease loci.

This study searched for putative regulatory mutations specific to the human lineage by looking for sequences that are highly conserved between chimpanzees and other species, but are not present in the human genome. The 500-odd human-specific deletions tend to lie in non-coding DNA stretches and near genes involved in steroid hormone signalling and neural function. This is illustrated with two examples, one of which affects penile anatomy whereas the other affects brain size.

Current aptamer discovery approaches are unable to probe the complete space of possible sequences. Here, the authors use machine learning to facilitate the development of DNA aptamers with improved binding affinities, and truncate them without significantly compromising binding affinity.

ChIP-Seq data are usually analyzed with approaches developed for microarrays, which only consider binding events within a few kilobases of a gene. McLean et al. present an algorithm that takes into account more distant events, thereby improving functional annotation of regulatory regions.

O’Connell et al. construct a new genome-wide imputation reference panel comprising 2,269 individuals of Sub-Saharan African ancestries. They adapt DeepVariant to create best practices for reference panel development and generate a high quality, publicly available resource that will further empower high resolution genome-wide imputation efforts in individuals of African ancestries.

Genome in a Bottle Consortium presents a high-confidence dataset for benchmarking small variant calls in human genomes.

DeepVariant uses convolutional neural networks to improve the accuracy of variant calling.