Auto-HMM-LMF

Auto-HMM-LMF performs feature selection and drug-response prediction by combining Hidden Markov Models (HMM) for mutation signature assignment, autoencoders for gene expression and copy number variation feature extraction, and logistic matrix factorization to model cell line–drug associations.

Key Features:

• HMM and Signature Assignments: Uses Hidden Markov Models to identify patterns in mutation data and assign signatures for mutation-based feature selection.
• Autoencoders: Applies autoencoders to gene expression and copy number variation (CNV) data to learn compressed representations and extract predictive features.
• Logistic Matrix Factorization: Applies logistic matrix factorization to selected features to predict drug response values and model interactions between cell lines and drugs.
• High-dimensional Feature Selection: Selects subsets of features from high-dimensional genomic data to improve predictive accuracy.
• Comparative Performance: Feature sets selected by Auto-HMM-LMF outperformed the ensemble feature selection method (EFS) in drug response prediction as reported.

Scientific Applications:

• Drug response prediction in cancer cell lines: Predicts cell line–drug response values from mutation, gene expression, and copy number variation data for pharmacogenomic analyses.
• Pharmacogenomics benchmarking: Validated on the Genomics of Drug Sensitivity in Cancer (GDSC) and the Cancer Cell Line Encyclopedia (CCLE) datasets for performance evaluation.
• Drug repositioning for head and neck cancer: Applied to identify potential candidate drugs for head and neck cancer.

Methodology:

Feature selection uses Hidden Markov Models with mutation signature assignment for mutation data and autoencoders for gene expression and copy number variation data; logistic matrix factorization is applied to the selected features to predict drug response values.