The power and limits of predicting inter-protein exon-exon interactions using protein 3D structures

Area of Science:

• Computational biology
• Structural biology
• Genomics

Background:

• Alternative splicing (AS) impacts cellular functions by altering protein-protein interactions (PPIs).
• Predicting AS-related PPI changes is limited by the scarcity of known exon-exon interactions (EEIs), covering less than 0.5% of human PPIs.
• There is a critical need to expand knowledge of EEIs for a deeper functional understanding of AS.

Purpose of the Study:

• To investigate the adaptability of existing 3D protein structure-based computational PPI interface prediction (PPIIP) methods for predicting EEIs.
• To assess the performance of PPIIP methods in predicting both residue-residue interactions (RRIs) and EEIs.

Main Methods:

• Utilized experimentally determined 3D structures of human protein heterodimers from the Protein Data Bank.
• Evaluated PPIIP methods on approximately 230,000 RRIs and 20,400 EEIs as ground truth.
• Applied computational pipelines to predict EEIs using PPIIP methods.

Main Results:

• Provided the first evidence that PPIIP methods can be adapted to predict EEIs.
• Achieved a performance score of up to approximately 76% (Area Under the ROC Curve) for EEI prediction.
• Established insights, data, and computational pipelines to guide future EEI prediction method development.

Conclusions:

• Existing PPIIP methods show significant adaptability for predicting EEIs, overcoming limitations of current approaches.
• This study lays the groundwork for enhanced computational tools to predict EEIs, crucial for understanding alternative splicing.
• The developed resources facilitate future research in computational prediction of exon-exon interactions and their functional implications.