Particle swarm optimization artificial intelligence technique for gene signature discovery in transcriptomic cohorts

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Personalized medicine relies on gene signatures, but few are clinically available for cancer.
• Current gene signature discovery often focuses on a single signature, limiting options.
• Multiple predictive signatures can be derived from the same dataset.

Purpose of the Study:

• To validate unique, biologically distinct gene signatures identified by Enhanced Binary Particle Swarm Optimization (EBPSO).
• To assess EBPSO's performance against Binary Particle Swarm Optimization (BPSO) in generating gene signatures for clinical translation.
• To explore the potential of EBPSO in overcoming limitations of traditional single gene signature discovery.

Main Methods:

• Utilized Enhanced Binary Particle Swarm Optimization (EBPSO), an advancement over Binary Particle Swarm Optimization (BPSO).
• Applied EBPSO to discover multiple, unique gene signatures from transcriptomics data.
• Validated the identified gene signatures on clinical transcriptomics cohorts.

Main Results:

• EBPSO achieved comparable accuracy to BPSO but with significantly smaller feature sets and faster runtimes.
• 100% accuracy was reached in most tested datasets.
• EBPSO successfully identified accurate, succinct, and prognostically significant gene signatures unique from each other.

Conclusions:

• EBPSO is a promising alternative for generating diverse and clinically relevant gene signatures.
• The identified signatures offer biological insights correlated with specific cancer types.
• EBPSO facilitates the selection and validation of gene signatures for clinical application in personalized cancer medicine.