Improving variant calling using population data and deep learning

Large-scale population variant data is often used to filter and aid interpretation of variant calls in a single sample. These approaches do not incorporate population information directly into the process of variant calling, and are often limited to filtering which trades recall for precision. In this study, we modify DeepVariant to add a new channel encoding population allele frequencies from the 1000 Genomes Project. We show that this model reduces variant calling errors, improving both precision and recall. We assess the impact of using population-specific or diverse reference panels. We achieve the greatest accuracy with diverse panels, suggesting that large, diverse panels are preferable to individual populations, even when the population matches sample ancestry. Finally, we show that this benefit generalizes to samples with different ancestry from the training data even when the ancestry is also excluded from the reference panel.