Leveraging Large Language Models for Cancer Variant Classification: A Comparative Study of GPT-4o, LLaMA 3, and Qwen 2.5

Abstract

Interpreting genomic variants from cancer sequencing data is a critical yet complex task in precision oncology. With advances in large language models (LLMs), there is increasing interest in leveraging their capacity for variant classification. This study benchmarks three state-of-the-art LLMs - GPT-4o, LLaMA 3, and Qwen 2.5 - on curated cancer variant databases to assess their utility in clinical genomic interpretation.

Keywords: CIViC; Cancer Variant Classification; Clinical Genomics; Genomic Profiling; Large Language Models; OncoKB; Precision Oncology.