Related Experiment Video
Updated: Jun 1, 2025
08:46
Implementation of In Vitro Drug Resistance Assays: Maximizing the Potential for Uncovering Clinically Relevant Resistance Mechanisms
Published on: December 9, 2015
10.5K
An ensemble machine learning-based performance evaluation identifies top In-Silico pathogenicity prediction methods
Subrata Das1, Vatsal Patel1, Shouvik Chakravarty1,2
1Biotechnology Research and Innovation Council-National Institute of Biomedical Genomics (BRIC-NIBMG), National Institute of Biomedical Genomics, Kalyani, West Bengal, India.
Biodata Mining
|January 20, 2025
Summary
This study developed an ensemble machine learning approach to rank cancer mutation pathogenicity scoring algorithms. The top-performing algorithms accurately distinguish driver mutations from passenger mutations, improving cancer gene prioritization.
Area of Science:
- Genomics
- Bioinformatics
- Computational Biology
Background:
- Accurate identification of cancer driver mutations is crucial for patient management.
- Existing mutation pathogenicity algorithms show significant performance variation, even for known drivers.
- Inconsistencies in mutation assessment hinder reliable clinical application.
Purpose of the Study:
- To develop an ensemble machine learning (ML) approach for evaluating pathogenic and conservation scoring algorithms (PCSAs).
- To rank PCSAs based on their ability to differentiate pathogenic driver mutations from benign passenger mutations in head and neck squamous cell carcinoma (HNSC).
- To identify the most effective PCSAs for prioritizing cancer-driving mutations.
Main Methods:
- Utilized a dataset of 502 HNSC patients with mutations classified across 299 high-confidence cancer driver genes.
- Annotated each mutation with 41 PCSAs.
- Employed logistic regression, random forest, and support vector machine ML algorithms with recursive feature elimination to rank PCSAs, using rank-average-sort and rank-sum-sort for final ranking.
Main Results:
- The random forest algorithm achieved the highest performance (AUC-ROC 0.89) in distinguishing driver from passenger mutations.
- The top 11 PCSAs, identified via rank-sum distribution, significantly outperformed the remaining 30 PCSAs (p < 2.22e-16).
- Top-ranked PCSAs demonstrated robust performance on validation cohorts across HNSC, breast, lung, and colorectal cancers.
Conclusions:
- An ensemble ML approach effectively evaluates and ranks PCSAs for differentiating pathogenic driver mutations from benign passenger mutations.
- Data-driven selection of PCSAs is essential, as some popular algorithms performed poorly.
- The validated top PCSAs offer a more reliable tool for cancer mutation analysis and prioritization across various cancer types.
