Dual-stage optimizer for systematic overestimation adjustment applied to multi-objective genetic algorithms for...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Genomics

Background:

• Biomarker panel selection from omics data is challenging due to high dimensionality and limited samples.
• Wrapper feature selection methods, like genetic algorithms, are used with machine learning for biomarker discovery.
• Existing methods often overestimate model performance, especially in multi-objective optimization.

Purpose of the Study:

• To address performance overestimation in multi-objective biomarker selection during optimization.
• To introduce a novel algorithm, Dual-stage Optimizer for Systematic overestimation Adjustment in Multi-Objective problems (DOSA-MO).
• To improve the selection of biomarker panels for enhanced predictive accuracy.

Main Methods:

• Developed DOSA-MO, a multi-objective optimization wrapper algorithm.
• DOSA-MO learns to predict and adjust for performance overestimation during optimization.
• Evaluated DOSA-MO by comparing it with a state-of-the-art genetic algorithm.

Main Results:

• DOSA-MO significantly improves the performance of genetic algorithms on external datasets.
• The algorithm enhances the accuracy of cancer subtype classification.
• Improved prediction of patient overall survival was observed.

Conclusions:

• DOSA-MO effectively reduces performance overestimation in multi-objective biomarker selection.
• The proposed method enhances the reliability and accuracy of biomarker panels identified from omics data.
• DOSA-MO offers a valuable advancement for machine learning applications in cancer research.