A comprehensive review of variant calling tools for RNA-seq: Challenges and advances

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• RNA sequencing (RNA-seq) is crucial for transcriptomic variation analysis.
• Detecting expressed genetic variants from RNA-seq presents unique challenges.
• These challenges include alternative splicing, allele-specific expression, and RNA editing.

Purpose of the Study:

• To review and compare computational strategies for RNA-seq variant calling.
• To highlight advancements from classical models to machine learning and deep learning.
• To emphasize the integration of these methods for improved accuracy and biological interpretation.

Main Methods:

• Classical statistical models (e.g., SAMtools, VarScan2, bcftools).
• Sophisticated tools with pair-Hidden Markov Models and de Bruijn graph assemblies (e.g., GATK HaplotypeCaller, Octopus).
• Machine learning (e.g., RNA-SNPhunter, RVboost) and deep learning (e.g., DeepVariant) approaches.

Main Results:

• Classical methods offer efficiency but limited sensitivity to RNA-specific artifacts.
• Advanced tools improve haplotype-aware detection across spliced regions.
• AI-driven methods enhance variant classification and learn complex error signatures.

Conclusions:

• The best RNA-seq variant calling pipelines integrate statistical rigor with AI adaptability.
• These methods balance precision and recall across variant types and expression levels.
• Methodological innovations are advancing RNA-seq variant calling for multi-omic integration and precision medicine.