A hybrid feature selection algorithm and its application in bioinformatics.

Area of Science:

• Bioinformatics and computational biology
• Machine learning applications
• Genomics and cancer research

Background:

• High-dimensional datasets necessitate advanced feature selection techniques for improved accuracy.
• Existing methods often struggle with the complexity and scale of modern biological data.
• The urgent need for effective feature selection is driven by the vast expansion of information, particularly in bioinformatics.

Purpose of the Study:

• To propose and evaluate the hybrid MMPSO method for optimal feature subset selection.
• To enhance classification accuracy using the MMPSO algorithm on diverse datasets.
• To identify potential diagnostic and prognostic biomarkers for liver hepatocellular carcinoma (LIHC).

Main Methods:

• Developed a hybrid MMPSO method combining feature ranking and heuristic search.
• Validated the MMPSO algorithm on ten datasets from the UCI Machine Learning Repository.
• Applied the MMPSO method to analyze gene expression data from The Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression (GTEx) for LIHC.

Main Results:

• The MMPSO algorithm demonstrated superior classification accuracy compared to other methods.
• Identified an 18-gene signature capable of distinguishing between normal and LIHC tumor samples.
• A seven-gene combination achieved an area under the curve (AUC) greater than 0.99 for tumor classification.
• Six identified genes showed significant correlation with patient survival time.

Conclusions:

• The MMPSO algorithm is effective for feature extraction from high-dimensional biological data.
• The identified 18-gene signature holds promise for LIHC diagnosis and prognosis.
• The MMPSO method provides new avenues for discovering biomarkers and therapeutic targets in cancer research.