AI/ML-Driven Gene Analysis: New Perspectives on Variant Calling in Normal Human Tissue Using scRNA-seq Data

Advances in artificial intelligence (AI) and machine learning (ML) have rapidly transformed bioinformatics, offering new solutions for data-intensive challenges in genomics. In this chapter, we introduce a practical protocol that contrasts a conventional variant-calling approach with a modern AI/ML-based method using single-cell RNA sequencing (scRNA-seq) data. Focusing on somatic variant detection in a normal human tissue, we explore the characteristics of variant calls-including variant types, read depth, and variant allele frequency (VAF)-and address technical challenges such as alignment artifacts near splice junctions and noise in RNA-seq data. We demonstrate that AI-based methods, such as DeepVariant, provide enhanced accuracy and confidence in genotype prediction compared to rule-based tools such as RNA-Mutect2. By outlining comparative workflows and analysis outputs, this chapter highlights the growing potential of AI/ML in improving variant discovery and interpretation in transcriptomic data, particularly in contexts lacking matched-normal controls.