Related Experiment Video
Updated: Oct 6, 2025
07:15
Determining the Likelihood of Variant Pathogenicity Using Amino Acid-level Signal-to-Noise Analysis of Genetic Variation
Published on: January 16, 2019
11.1K
LYRUS: a machine learning model for predicting the pathogenicity of missense variants
Jiaying Lai1,2, Jordan Yang3, Ece D Gamsiz Uzun2,4,5
1Center for Biomedical Informatics, Brown University, Providence, RI 02903, USA.
Bioinformatics Advances
|January 17, 2022
Summary
This study introduces LYRUS, a machine learning tool to predict pathogenic single amino acid variations (SAVs). LYRUS utilizes sequence, structure, and dynamics features, including a novel co-evolution metric, to enhance variant effect prediction.
Area of Science:
- Genomics and Bioinformatics
- Computational Biology
- Molecular Genetics
Background:
- Single amino acid variations (SAVs) are a major source of human genetic diversity.
- Identifying pathogenic SAVs is crucial for understanding complex disease genetics.
- Existing prediction methods primarily rely on sequence or structural data.
Purpose of the Study:
- To develop a novel machine learning method for predicting the pathogenicity of SAVs.
- To incorporate diverse features, including sequence, structure, dynamics, and co-evolution information.
- To provide an accessible tool for variant effect prediction.
Main Methods:
- Developed LYRUS, a machine learning model using an XGBoost classifier.
- Integrated five sequence-based, six structure-based, and four dynamics-based features.
- Introduced a novel sequence co-evolution feature: the variation number.
Main Results:
- LYRUS was trained on 22,639 SAVs from 4363 protein structures (ClinVar database).
- Performance was evaluated using the VariBench testing dataset.
- LYRUS demonstrated comparable performance to existing variant effect predictors and was benchmarked against Deep Mutational Scanning datasets.
Conclusions:
- LYRUS offers a robust approach to predicting SAV pathogenicity by integrating multiple feature types.
- The inclusion of the variation number feature enhances predictive capabilities.
- The tool is freely available, facilitating further research in variant effect prediction.
