Semi-supervised Learning Predicts Approximately One Third of the Alternative Splicing Isoforms as Functional Proteins

Area of Science:

• Molecular Biology
• Genomics
• Bioinformatics

Background:

• Alternative splicing is a widespread biological process affecting most human multi-exon genes.
• The precise impact of alternative splicing on the production of stable, functional proteins remains largely unknown due to experimental limitations and lack of negative data.

Purpose of the Study:

• To develop and validate a novel computational method for predicting protein-coding alternative splicing events.
• To estimate the proportion of alternative splicing events that generate stable protein isoforms.
• To explore the functional and structural implications of alternatively spliced proteins.

Main Methods:

• Development of a semi-supervised machine learning algorithm named Positive Unlabeled Learning for Splicing Elucidation (PULSE).
• PULSE utilizes 48 diverse features to analyze splicing events.
• Algorithm validation using known protein isoforms and mass spectrometry (MS) data, achieving an AU-ROC of 0.85.

Main Results:

• Prediction that approximately 32% of "exon skipping" alternative splicing events result in stable proteins.
• Identification of a substantial number of potentially novel protein isoforms arising from alternative splicing.
• Analysis of the distribution of these protein isoforms across different functional categories and their structural consequences.

Conclusions:

• Alternative splicing contributes significantly to proteomic diversity, generating a large repertoire of stable, potentially novel proteins.
• The PULSE algorithm provides a robust tool for investigating alternative splicing and its downstream effects on protein expression.
• Further research into alternatively spliced proteins is warranted to fully understand their functional significance and structural impact.