AI Algorithms Improve Genetic Mutation Detection in Cancer Diagnostics

Accurately identifying genetic mutations is central to cancer diagnostics and genomic research, but current methods struggle with complex sequencing data and limited clinical samples. Tumor analysis often requires matched healthy tissue for comparison, which is not always available, while RNA sequencing is complicated by biological editing and technical noise. These challenges can delay diagnosis and limit the clinical utility of genomic data. Researchers have now developed artificial intelligence (AI) tools that overcome these barriers, enabling more precise mutation detection directly from long-read DNA and RNA data.

Researchers at The University of Hong Kong (Hong Kong, China) have developed two deep-learning algorithms, ClairS-TO and Clair3-RNA, designed specifically for long-read sequencing technologies, which capture extended stretches of DNA or RNA and offer richer genetic information than short-read methods. ClairS-TO focuses on tumor-only DNA analysis, using a dual-network architecture that distinguishes true cancer mutations from sequencing errors without requiring matched normal tissue.

Image: Long-read DNA and RNA sequencing data analyzed using AI algorithms can improve mutation detection accuracy (Photo courtesy of 123RF)

Clair3-RNA is the first deep-learning-based small variant caller optimized for long-read RNA sequencing, enabling accurate differentiation between genuine genetic variants, RNA editing events, and technical artifacts. Both algorithms were evaluated on complex sequencing datasets and demonstrated significantly improved accuracy compared with existing approaches.

ClairS-TO reliably detected somatic mutations in tumor samples alone, addressing a major limitation in cancer diagnostics where normal tissue samples are unavailable or impractical to obtain. Clair3-RNA achieved high precision in identifying small variants directly from RNA sequencing data while accounting for RNA editing and noise. The findings from both tools were published in Nature Communications, confirming their robustness for both clinical and research applications.

These advances expand the practical use of long-read sequencing in precision medicine by reducing costs, simplifying workflows, and improving reliability. ClairS-TO enables broader access to accurate tumor profiling, while Clair3-RNA allows simultaneous analysis of gene expression and genetic variation from a single RNA dataset.

The algorithms are part of the open-source Clair series, which is already widely adopted globally. Future applications include improved cancer diagnosis, personalized treatment selection, and deeper insights into RNA biology and disease mechanisms.

“ClairS-TO and Clair3-RNA, along with other algorithms in the Clair series, have established a solid foundation for deep-learning-driven genetic mutation discovery and accelerated the adoption of precision medicine and clinical genomics,” said Professor Ruibang Luo, lead investigator of the study.

Related Links:
The University of Hong Kong