Extreme Gradient Boosting Tuned with Metaheuristic Algorithms for Predicting Myeloid NGS Onco-Somatic Variant...

Area of Science:

• Bioinformatics
• Genomics
• Onco-somatic Genetics

Background:

• Next-generation sequencing (NGS) generates vast amounts of data, complicating variant interpretation in cancer genetics.
• Machine learning, specifically Extreme Gradient Boosting (XGBoost), offers powerful solutions for analyzing complex genomic datasets.
• The myeloid panel is crucial for diagnosing and treating myeloid neoplasms by identifying specific genetic mutations.

Purpose of the Study:

• To develop and optimize an XGBoost-based machine learning tool for predicting mutation pathogenicity in the myeloid panel.
• To enhance the accuracy of variant interpretation in onco-somatic genetics using NGS data.
• To improve the diagnostic and therapeutic strategies for myeloid neoplasms.

Main Methods:

• Utilized datasets from myeloid panel NGS analysis, comprising 15,977 variants (SNVs, MNVs, INDELs), to train the XGBoost algorithm.
• Optimized XGBoost hyperparameters using metaheuristic algorithms, specifically Differential Evolution (DE).
• Compared the tool's predictions against human expert decisions and other prediction tools.

Main Results:

• Achieved high performance metrics: 99.35% accuracy, 98.70% precision, 98.71% specificity, and 100% sensitivity.
• The optimized XGBoost model demonstrated superior performance in predicting mutation pathogenicity.
• The tool's predictions closely aligned with expert evaluations.

Conclusions:

• The developed XGBoost tool effectively predicts mutation pathogenicity in the myeloid panel, aiding in cancer diagnosis.
• Optimizing machine learning models with metaheuristic algorithms significantly enhances predictive accuracy for genomic data.
• This approach offers a valuable advancement for precision medicine in treating myeloid neoplasms.