CancerVar

CancerVar provides automated evidence-based clinical interpretation of somatic mutations in cancer, extending interpretation beyond manually curated hotspot databases such as CIViC and OncoKB.

Key Features:

• Comprehensive Mutation Scoring: Evaluates over 12.9 million somatic mutations and classifies them into four tiers: Strong Clinical Significance, Potential Clinical Significance, Uncertain Clinical Significance, and Benign/Likely Benign.
• Guideline-Based Classification: Applies the AMP/ASCO/CAP 2017 guidelines as the basis for initial rule-based classification incorporating functional and clinical evidence.
• Deep Learning Integration: Uses a semi-supervised generative adversarial network (SGAN) to predict oncogenicity by integrating functional and clinical data.
• Training and Validation: Trains the SGAN model on an in-house database of 5,234 somatic mutations and validates performance against 6,226 variants curated from literature.
• Independent Dataset Comparison: Compares predictions with several independent datasets to assess utility for variants with previously unknown interpretations.
• Automated Text Generation: Generates summarized descriptive interpretations for hotspot mutations including diagnostic, prognostic, targeted drug response, and clinical trial information.

Scientific Applications:

• Clinical interpretation: Supports evidence-based classification of somatic variants for clinical significance assessment in oncology.
• Hypothesis generation: Facilitates generation of research hypotheses about functional and clinical impacts of novel and known mutations.
• Personalized medicine: Aids researchers and clinicians in evaluating diagnostic, prognostic, and targeted therapy implications of somatic mutations to inform personalized oncology approaches.

Methodology:

Combines AMP/ASCO/CAP 2017 guideline-based rule scoring with a semi-supervised generative adversarial network (SGAN) trained on 5,234 somatic mutations and validated against 6,226 curated variants, and compares predictions to independent datasets.