Evolving Multiobjective Cancer Subtype Diagnosis From Cancer Gene Expression Data

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning in Oncology

Background:

• Accurate cancer detection and diagnosis are critical for early prevention and effective treatment strategies.
• Existing methods for cancer subtype diagnosis often exhibit limitations in diagnostic ability and generalization.
• The need for robust and highly accurate diagnostic tools in oncology is paramount.

Purpose of the Study:

• To develop a multiobjective PSO-based hybrid algorithm (MOPSOHA) for optimizing cancer data diagnosis.
• To simultaneously enhance classification accuracy, reduce feature numbers, and manage relevance and redundancy.
• To achieve high classification power and robustness in cancer subtype diagnosis.

Main Methods:

• Implementation of a novel binary encoding strategy for selecting informative gene subsets.
• Integration of a mutation operator to improve the exploration capabilities of the particle swarm optimization (PSO).
• Application of a local search method using differential evolution's (DE) 'best/1' mutation for exploiting promising solution areas.

Main Results:

• MOPSOHA demonstrated superior performance across 41 diverse cancer datasets, including gene expression and independent disease datasets.
• The algorithm achieved high classification power and robustness compared to existing state-of-the-art methods.
• Simultaneous optimization of multiple objectives (feature count, accuracy, relevance, redundancy) proved effective.

Conclusions:

• The proposed MOPSOHA algorithm offers a significant advancement in cancer subtype diagnosis.
• Its hybrid approach effectively balances multiple optimization objectives for improved diagnostic outcomes.
• MOPSOHA shows strong potential for clinical application in cancer diagnostics due to its superior performance and generalization.