IsoFrog: a reversible jump Markov Chain Monte Carlo feature selection-based method for predicting isoform functions.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Alternative splicing generates multiple functional isoforms from a single gene.
• Predicting isoform functions is crucial for understanding gene roles.
• Existing machine learning methods often struggle with feature relevance and noise.

Purpose of the Study:

• To develop an improved machine learning approach for gene isoform function prediction.
• To address the challenge of irrelevant features in existing prediction models.
• To enhance model accuracy by incorporating feature selection.

Main Methods:

• Developed IsoFrog, a feature selection-based approach for isoform function prediction.
• Utilized a reversible jump Markov Chain Monte Carlo (RJMCMC) framework for feature importance assessment.
• Employed sequential feature selection to identify function-relevant features and a modified domain-invariant partial least squares (diPLS) for prediction.

Main Results:

• IsoFrog demonstrated superior performance compared to six state-of-the-art methods across three datasets.
• The RJMCMC-based feature selection component outperformed three traditional feature selection methods.
• The method effectively screens relevant features while eliminating noise, enhancing prediction accuracy.

Conclusions:

• The proposed IsoFrog methodology significantly improves gene isoform function prediction.
• Feature selection is a viable strategy to enhance machine learning models in this domain.
• This work is expected to advance isoform function identification and inspire new computational methods.