LYRUS

LYRUS predicts the pathogenicity of single amino acid variations (SAVs) by integrating sequence, structure, and dynamics information with machine learning to assess variant impact.

Key Features:

• Machine Learning Framework: Employs an XGBoost classifier selected via the automated machine learning tool TPOT.
• Feature Integration: Combines five sequence-based features, six structure-based features, and four dynamics-based features.
• Sequence Co-evolution Feature: Introduces a "variation number" metric to capture sequence co-evolution that distinguishes pathogenic from neutral SAVs.
• Comprehensive Evaluation: Evaluated on a ClinVar-derived dataset of 4,363 protein structures associated with 20,307 SAVs, demonstrating superior accuracy, specificity, F-measure, and Matthews correlation coefficient compared to PolyPhen2, PROVEAN, SIFT, Rhapsody, EVMutation, MutationAssessor, SuSPect, FATHMM, and MVP.
• Application to Cancer Genes: Applied to variants in PTEN and TP53 to assess performance on cancer-related SAVs.

Scientific Applications:

• Pathogenic SAV Identification: Prioritizing and classifying single amino acid variants for downstream genomic studies.
• Cancer Variant Analysis: Assessing the impact of variants in cancer-associated genes such as PTEN and TP53.
• Benchmarking Predictors: Comparative evaluation and benchmarking of pathogenicity prediction methods using ClinVar-derived variants.

Methodology:

Uses an XGBoost classifier selected by TPOT; integrates five sequence-based, six structure-based, and four dynamics-based features including a "variation number" co-evolution metric; evaluated on 4,363 protein structures comprising 20,307 SAVs from ClinVar against multiple established predictors.