SLiMEnrich: computational assessment of protein–protein interaction data as a source of domain-motif interactions

Idrees S, Pérez-Bercoff Å & Edwards RJ. (2018) SLiMEnrich: computational assessment of protein–protein interaction data as a source of domain-motif interactions. PeerJ 6:e5858 https://doi.org/10.7717/peerj.5858

Abstract

Many important cellular processes involve protein–protein interactions (PPIs) mediated by a Short Linear Motif (SLiM) in one protein interacting with a globular domain in another. Despite their significance, these domain-motif interactions (DMIs) are typically low affinity, which makes them challenging to identify by classical experimental approaches, such as affinity pulldown mass spectrometry (AP-MS) and yeast two-hybrid (Y2H). DMIs are generally underrepresented in PPI networks as a result. A number of computational methods now exist to predict SLiMs and/or DMIs from experimental interaction data but it is yet to be established how effective different PPI detection methods are for capturing these low affinity SLiM-mediated interactions. Here, we introduce a new computational pipeline (SLiMEnrich) to assess how well a given source of PPI data captures DMIs and thus, by inference, how useful that data should be for SLiM discovery. SLiMEnrich interrogates a PPI network for pairs of interacting proteins in which the first protein is known or predicted to interact with the second protein via a DMI. Permutation tests compare the number of known/predicted DMIs to the expected distribution if the two sets of proteins are randomly associated. This provides an estimate of DMI enrichment within the data and the false positive rate for individual DMIs. As a case study, we detect significant DMI enrichment in a high-throughput Y2H human PPI study. SLiMEnrich analysis supports Y2H data as a source of DMIs and highlights the high false positive rates associated with naïve DMI prediction. SLiMEnrich is available as an R Shiny app. The code is open source and available via a GNU GPL v3 license at: https://github.com/slimsuite/SLiMEnrich. A web server is available at: http://shiny.slimsuite.unsw.edu.au/SLiMEnrich/.

Evaluation of protein-protein interaction detection methods as a source of capturing domain-motif interactions

Sobia Idrees, Richard J Edwards

This work was presented at the Sydney Bioinformatics Research Symposium 2018. (Abstract below.) Click on thumbnail for full resolution PDF.

Abstract

One of the main pursuits in proteomics is to understand the complex network of protein-protein Interactions (PPI) that underpin biological processes. Two major classes of PPI are domain-domain interactions (DDI) between globular proteins, and domain-motif interactions (DMI) between a globular domain and a short linear motif (SLiM) in its partner. Advances in high-throughput experimental techniques have been applied at large-scale in an attempt to characterise the interactome of various organisms. However, PPI networks being identified by these high-throughput experiments have low resolution as compared to low-throughput technologies, such as protein co-crystallization. Furthermore, large-scale approaches may be poor at capturing low affinity or transient interactions, which includes the majority of known DMI. To date, several studies have been conducted to identify how well these PPI data can capture protein complexes, but the ability of high-throughput PPI-detection methods to capture DMI remains a largely unanswered question.

To help system biologists choose appropriate methods for predicting different types of interactions, we conducted a comprehensive comparison study on existing high-throughput PPI datasets. We have integrated PPI data, SLiM predictions, domain compositions and known SLiM-domain binding partnerships to identify possible DMI and DDI within interactomes. We identify PPI data that are enriched for DMI or DDI versus a background expectation generated by randomising the PPI within the network. Despite returning relatively few experimentally validated DMI when compared to interaction databases, we present evidence that high-throughput PPI data is enriched for DMI and thus potentially useful for the prediction of novel SLiMs. We discuss the relative merits of co-fractionation followed by mass spectrometry (CoFrac-MS), affinity purification coupled mass spectrometry (AP-MS), and yeast two hybrid (Y2H) for capturing DMI and DDI, as well as potential quality versus quantity trade-offs in DMI prediction.

The SLiMEnrich Shiny App is now live

Sobia’s first Shiny App is now up and running for final pre-publication testing on our new EdwardsLab RShiny server. Please feel free to try it out. Any comments and suggestions can be posted here, by email, or via the GitHub issues page.

The SLiMEnrich App allows users to predict Domain-Motif Interactions (DMIs) from Protein-Protein Interaction data (PPI) and to estimate the background distribution of expected DMI by chance through a randomisation approach. The walkthrough has more details.

SLiMEnrich can be used to:

Estimate the enrichment of SLiM-mediated DMI in a given PPI dataset.

Generate predictions for DMI-mediated from a PPI dataset. Predictions are based on known SLiM-Domain interactions.

Estimate the False Positive Rate of those DMI predictions.

With a bit of imagination, SLiMEnrich can be adapted to generate and assess predictions for other kinds of interactions. See docs for details.

SLiMEnrich is available via the SLiMEnrich RShiny webserver and can be downloaded for local running from the SLiMEnrich GitHub repo.