CrossIsoFun: predicting isoform functions using the integration of multi-omics data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene isoforms, arising from alternative splicing, can possess distinct biological functions.
• Understanding isoform-level functions is crucial for deciphering genomic functional diversity.
• Current computational methods struggle with integrating heterogeneous multi-omics data for isoform function prediction, particularly underutilizing isoform-isoform interactions (IIIs).

Purpose of the Study:

• To develop an advanced computational framework for accurate isoform function prediction.
• To enhance multi-omics data integration for improved functional annotation at the isoform level.
• To effectively incorporate isoform-isoform interactions (IIIs) into prediction models.

Main Methods:

• Introduction of CrossIsoFun, a multi-omics data analysis framework.
• Utilizing graph convolutional networks (GCNs) for omics-specific classification.
• Employing a View Correlation Discovery Network (VCDN) for cross-omics integration.
• Generating synthetic IIIs from protein-protein interactions (PPIs) using a cycle-consistency generative adversarial network (cycleGAN) to enrich interactomics data.

Main Results:

• CrossIsoFun demonstrates superior performance compared to state-of-the-art methods on multiple datasets.
• The framework effectively integrates mRNA expression, sequence, and PPI data.
• Predictions generated by CrossIsoFun show consistency with subcellular localization and literature-supported isoform annotations.

Conclusions:

• CrossIsoFun offers a robust and effective approach for isoform function prediction by integrating diverse omics data and IIIs.
• The framework addresses limitations in current multi-omics integration methods.
• This work advances the computational prediction of isoform functions, providing valuable insights into genomic functional diversity.