Complementary feature selection from alternative splicing events and gene expression for phenotype prediction.

Area of Science:

• Bioinformatics and Computational Biology
• Genomics and Transcriptomics
• Machine Learning in Medicine

Background:

• Predicting medical prognoses from biomarker patterns is crucial in bioinformatics.
• Current methods often rely on gene expression from RNA sequencing (RNA-Seq) data.
• Alternative splicing generates numerous isoforms, presenting a challenge and opportunity for phenotype prediction.

Purpose of the Study:

• To identify optimal methods for transcript quantification, feature engineering, and filtering for phenotype prediction using RNA-Seq data.
• To analyze the complementary nature of gene and isoform expression data.
• To assess the feasibility of using isoforms as biomarker candidates.

Main Methods:

• Empirical evaluation of transcript quantification and feature selection pipelines.
• Development and assessment of a univariate filtering algorithm for feature selection.
• Cross-validation prediction tests using datasets from non-small cell lung cancer (NSCLC), chronic obstructive pulmonary disease (COPD), and amyotrophic lateral sclerosis (ALS) models.

Main Results:

• Isoform features provide complementary, non-redundant information that enhances predictive power when prioritized and filtered.
• A univariate filtering approach effectively selects top-ranking features for phenotype prediction.
• The study systematically compared different isoform quantification pipelines.

Conclusions:

• Optimized methods for transcript quantification and feature selection improve phenotype prediction accuracy.
• Isoform data offers valuable, complementary insights beyond gene expression.
• This work facilitates the identification of novel isoform-based biomarkers for medical prognoses.