G.AI: an AI-driven platform for phenotype standardization, variant interpretation and structured clinical reporting...

The diagnosis of rare diseases increasingly relies on the interpretation of high-throughput next-generation sequencing (NGS) data. As sequencing volume expands, the analytical burden grows substantially, and manual workflows become increasingly difficult to scale and prone to inconsistency. To address these challenges, we developed G.AI, an interpretable and traceable artificial intelligence (AI)-assisted genomic analysis platform that integrates automated phenotype standardization, variant pathogenicity ranking, and structured clinical reporting.

The platform uses a modular architecture comprising data parsing, AI-driven inference, and structured report generation. Performance was assessed using 39,156 multicenter whole-exome sequencing (WES)/ parent-child trio sequencing (WES Trio) cases from China, including 7,097 confirmed pathogenic/likely pathogenic (P/LP) single-nucleotide variants (SNVs) positive cases. Key evaluation metrics included phenotype-model concordance, Top-1, Top-3 and Top-20 variant pathogenicity ranking accuracy and workflow efficiency.

The AI-Human Phenotype Ontology (HPO) phenotype standardization model achieved 94% concordance with manual review. The pathogenicity-ranking model reached Top-1 95%, Top-3 98%, and Top-20 99.6% accuracy among positive cases, with metabolic disorders achieving 100% Top-3 accuracy. Additional analysis on non-diagnostic cases demonstrated low false prioritization rates and good model specificity. Total analysis time decreased from 4 to 6 h to 48 ± 12 min, demonstrating a significant improvement in efficiency.

By integrating automated phenotype processing, variant annotation, and AI-driven pathogenicity evaluation, G.AI substantially enhances the accuracy, consistency, and scalability of rare disease variant interpretation. Its transparent and traceable workflow provides a robust foundation for large-scale clinical genomic applications.