A Global Assessment of the Transcription-Dependent Single Nucleotide Variants Relies on the Characteristics of RNA-Sequencing Technologies

Single nucleotide variants (SNVs) are crucial in cancer occurrence and development. SNVs at the transcriptomic level generally come from genomic variants (g-tSNVs) and RNA editing (e-tSNVs). The types and quantities of e-tSNVs remain a subject of debate due to a relatively poor understanding of RNA editing processes. Herein, we developed TSCS (Transcript SNVs Classifier relying on complementary sequencings), a machine learning classifier that integrates short-read (MGI) and long-read (PacBio) RNA-seq data to accurately distinguish true transcript SNVs using stringent criteria. Applied to five colorectal cancer cell lines (HCT15, LoVo, SW480, SW620, and HCT116), TSCS demonstrated superior accuracy and sensitivity, outperforming established tools (GATK, BCFtools, Longshot, RED_ML). It increased the total detected transcript SNVs by 31.83% on average, with g-tSNVs and e-tSNVs exceeding conventional methods by >1-fold and >2-fold, respectively. TSCS achieved mean recall rates of 75.3% for g-tSNVs and 77.2% for e-tSNVs. Notably, for the first time, e-tSNVs were found in a relatively large proportion of total transcript SNVs in cancer cell lines, approximately 40%. Of the identified e-tSNVs, 80% were attributed to the known RNA editing, but the other e-tSNVs did not fall into any known category. Importantly, the e-tSNVs uniquely detected in this study showed distinct patterns in SNV types and genomic locations. Additionally, the transcript SNVs called by TSCS were partially confirmed using experimental approaches, such as Sanger sequencing, RNC-seq, and mass spectrometry. This study lays the foundation for surveying and appraising the cancer-related e-tSNVs.