Combining multiple hypothesis testing and affinity propagation clustering leads to accurate, robust and sample size...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Feature selection in gene expression data needs to be platform, disease, and dataset size independent.
• Hypothesis: Statistically significant genes contain clusters with shared biological functions relevant to disease.
• Proposes using gene cluster exemplars instead of a fixed number of top-ranked genes.

Purpose of the Study:

• To develop and evaluate a hybrid feature selection method (mAP-KL) for identifying informative gene subsets from microarray data.
• To improve the accuracy and biological relevance of gene signatures for disease classification.
• To create a data-driven, classifier-independent method applicable to various diseases and dataset sizes.

Main Methods:

• Combines multiple hypothesis testing with affinity propagation (AP) clustering.
• Utilizes the Krzanowski & Lai cluster quality index.
• Applied to real and simulated microarray data for performance comparison.

Main Results:

• mAP-KL demonstrated competitive classification results across various diseases and sample sizes.
• Achieved a high Area Under the Curve (AUC) score of 0.91 in neuromuscular diseases.
• Generated concise, biologically relevant gene expression signatures for diagnostic/prognostic use.

Conclusions:

• mAP-KL is a data-driven, classifier-independent hybrid feature selection method.
• Applicable to any disease classification problem using microarray data, irrespective of sample size.
• Effectively classifies samples from both small and large cohorts with high accuracy.