Determining rewiring effects of alternatively spliced isoforms on protein-protein interactions using a computational approach

The critical role of alternative splicing (AS) in cell functioning has recently become apparent, whether in studying tissue-or cell-specific regulation, or understanding molecular mechanisms governing a complex disorder. Studying the rewiring, or edgetic, effects of alternatively spliced isoforms on protein interactome can provide system-wide insights into these questions. Unfortunately, high-throughput experiments for such studies are expensive and time-consuming, hence the need to develop an in-silico approach. Here, we formulated the problem of characterization the edgetic effects of AS on protein-protein interactions (PPIs) as a binary classification problem and introduced a first computational approach to solve it. We first developed a supervised feature-based classifier that benefited from the traditional features describing a PPI, the problem-specific features that characterized the difference between the reference and alternative isoforms, and a novel domain interaction potential that allowed pinpointing the domains employed during a specific PPI. We then expanded this approach by including a large set of unlabeled interactomics data and developing a semi-supervised learning method. Our method called AS-IN (Alternatively Splicing INteraction prediction) Tool was compared with the state-of-the-art PPI prediction tools and showed a superior performance, achieving 0.92 in precision and recall. We demonstrated the utility of AS-IN Tool by applying it to the transcriptomic data obtained from the brain and liver tissues of a healthy mouse and western diet fed mouse that developed type two diabetes. We showed that the edgetic effects of differentially expressed transcripts associated with the disease condition are system-wide and unlikely to be detected by looking only at the gene-specific expression levels.