MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202641110456 A) filed by R Manasa on September 15, 2026, for Brca-Compass.
Inventors include R Manasa; D Jagadeesan; and Lokesh Ravi.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: The present invention provides a computer-implemented method, system, and non-transitory computer-readable medium for automated, explainable classification of BRCA1 gene variants as pathogenic or benign, structured prioritization of variants of uncertain significance, and generation of patient-level clinical decision-support output. A labelled variant cohort is derived from ClinVar through sequential filtering, with variants of uncertain significance permanently separated before any model- development step. A curated representation of 48 features, drawn from population-frequency, conservation, functional-prediction, protein-domain, and variant- consequence categories, is generated for each variant. A gradient-boosted decision-tree classifier, selected through comparative evaluation against CatBoost, LightGBM, Random Forest, Extra Trees, and a weighted ensemble, is trained to produce a calibrated pathogenicity probability, achieving a test-set ROC-AUC of 0.9973 and a Matthews Correlation Coefficient of 0.9538. An explainability module computes exact SHAP-based feature attributions for each prediction, which are automatically mapped to ACMG/AMP evidence codes comprising PVS1, PM1, PP3, BS1, and BP4. Predictions are further categorised into high-, medium-, or low- confidence tiers, and the classifier is validated against an independent saturation genome editing dataset without parameter refitting, while feature-category ablation confirms that its performance is not attributable to any single feature. The classifier is applied to a held-out cohort of 2,309 ClinVar-annotated variants of uncertain significance, assigning each variant to one of six ordered prioritization tiers and thereby reclassifying 276 variants as high-risk pathogenic, 28 as medium-risk pathogenic, and 24 as low-risk pathogenic, for a total of 328 variants of uncertain significance resolved toward a pathogenic classification, while 1,914 variants are resolved as high-confidence benign, and a patient-level clinical explanation module integrates the classification, feature attributions, evidence codes, and confidence tier into a structured, disclaimer-accompanied clinical narrative with recommended next steps, supporting transparent and clinically interpretable variant-pathogenicity decision support.
Disclaimer: Curated by HT Syndication.
