Small variant calling in the era of long-read sequencing.

Review: deep learning methods combined with alignment-based, assembly-based and pangenome-based strategies

This Review covers small variant calling, the predominant source of human genetic diversity, in the era of long-read sequencing. The distinct error profiles of long reads have shifted tools towards deep learning-based methods, reshaping the calling of germline, de novo, somatic and mosaic variants from both DNA and RNA. These frameworks are being developed in combination with alignment-based, assembly-based and pangenome-based strategies. The authors chart the trajectory towards complete telomere-to-telomere profiling of small variants and consider the computational innovations needed to benchmark newly accessible repetitive regions and complex variants.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

A narrative review without original data, useful as a map of the field for a laboratory considering long-read sequencing. The point that directly concerns diagnostics is benchmarking: as long as repetitive regions and complex variants lack a truth set, the claimed performance of callers in these regions cannot be verified, which limits the method validation required in accredited laboratories.

Analysis by Dr Thibaut Benquey