Clair3-RNA: a deep learning-based small variant caller for long-read RNA sequencing data

Abstract

Variant calling with long-read RNA sequencing (lrRNA-seq) helps to analyze full-length isoforms and gene expression but is complicated by high error rates, transcript diversity, RNA editing events, etc. Here, we propose Clair3-RNA, the first deep learning-based variant caller tailored for lrRNA-seq data. Building upon Clair series’ pipelines, Clair3-RNA enhances lrRNA-seq variant calling using optimized techniques, such as uneven coverage normalization, refined training data, editing site discovery, and haplotype phasing to enhance performance. Clair3-RNA supports various platforms, including PacBio, ONT complementary DNA sequencing (cDNA), and ONT direct RNA sequencing (dRNA). Clair3-RNA achieved a ~ 91% SNP F1-score on the ONT platform using the latest ONT SQK-RNA004 kit (dRNA004) and a ~ 92% SNP F1-score in PacBio Iso-Seq and MAS-Seq for variants with at least 4x coverage. With least 10x coverage and disregarding zygosity, the performance reached a ~ 95% and ~96% F1-score for ONT and PacBio, respectively. After phasing, the performance reached ~97% for ONT and ~98% for PacBio. Across GIAB samples, Clair3-RNA consistently outperformed existing callers and accurately identified RNA editing sites. Clair3-RNA is open-source at (https://github.com/HKU-BAL/Clair3-RNA).