GeneScope Predicting DNA Mutation Effects

Interpreting the functional impact of genetic variants plays a critical role in modern genomics, clinical diagnostics, and precision medicine. With the rapid expansion of wholegenome and exome sequencing, millions of Single Nucleotide Variants (SNVs) are discovered, yet only a small fraction has experimentally validated clinical labels. Traditional methods for variant effect prediction rely on handcrafted biological features or evolutionary conservation scores, which struggle to generalize across diverse genomic regions. This research investigates the use of Evo2, a transformer-based biological Large Language Model trained on 9.3 trillion DNA tokens for predicting the pathogenicity of SNVs. The system analyzes both wild-type and mutated sequences to capture contextual and evolutionary disruptions introduced by a mutation. A full prediction pipeline is developed that integrates reference genome extraction, mutation encoding, Evo2 inference, and comparison with ClinVar annotations. The model is evaluated using accuracy, precision, recall, F1-score, ROC-AUC, and ClinVar concordance. Results demonstrate that Evo2 provides highly competitive variant effect predictions, with strong discrimination between pathogenic and benign mutations. This work highlights the potential of large DNA language models as practical tools for genomic interpretation and presents a unified, accessible platform for variant analysis.