[Paper Review] SPRINT: Ultrafast protein-protein interaction prediction of the entire human interactome

Yiwei Li, Lucian Ilie|arXiv (Cornell University)|May 19, 2017

Bioinformatics and Genomic Networks35 references4 citations

TL;DR

SPRINT is a novel, ultrafast sequence-based algorithm for predicting protein-protein interactions (PPIs) across the entire human interactome. It uses a spaced-seed approach to identify conserved sequence motifs and propagates interaction scores through shared similarities, achieving state-of-the-art accuracy while running 100–1,000x faster and using minimal memory—making full interactome prediction feasible in under 2 hours on a standard server.

ABSTRACT

Proteins perform their functions usually by interacting with other proteins. Predicting which proteins interact is a fundamental problem. Experimental methods are slow, expensive, and have a high rate of error. Many computational methods have been proposed among which sequence-based ones are very promising. However, so far no such method is able to predict effectively the entire human interactome: they require too much time or memory. We present SPRINT (Scoring PRotein INTeractions), a new sequence-based algorithm and tool for predicting protein-protein interactions. We comprehensively compare SPRINT with state-of-the-art programs on seven most reliable human PPI datasets and show that it is more accurate while running orders of magnitude faster and using very little memory. SPRINT is the only program that can predict the entire human interactome. Our goal is to transform the very challenging problem of predicting the entire human interactome into a routine task. The source code of SPRINT is freely available from github.com/lucian-ilie/SPRINT/ and the datasets and predicted PPIs from www.csd.uwo.ca/faculty/ilie/SPRINT/.

Motivation & Objective

• To address the critical challenge of predicting all protein-protein interactions in the human interactome, a task previously infeasible due to computational cost.
• To develop a sequence-based PPI prediction method that is both highly accurate and scalable to whole proteomes.
• To eliminate bias in evaluation by using Park and Marcotte’s C1–C3 dataset scheme, which separates easy (C1) from hard-to-predict (C2, C3) interactions.
• To outperform existing methods in both accuracy and efficiency, particularly on the most difficult, biologically relevant interaction types (C2 and C3).
• To make whole-interactome PPI prediction a routine computational task by drastically reducing time and memory requirements.

Proposed method

• SPRINT uses a set of optimized spaced seeds to identify conserved k-mer subsequences across protein sequences, enabling efficient detection of potential interaction motifs.
• It builds hash tables to index all spaced-mers from the human proteome, enabling fast lookup and similarity computation.
• Similarity propagation is performed by identifying shared spaced-mers between interacting protein pairs and propagating their scores to other protein pairs.
• The score for each protein pair (Q₁, Q₂) is computed using a weighted formula that accounts for the number of supporting k-mer matches and the lengths of the proteins, as defined in Equation (3).
• Low-complexity or over-represented regions are filtered out to reduce false positives and improve biological relevance.
• Finally, all protein pairs are ranked by their cumulative score, with top-scoring pairs reported as predicted interactions.

Experimental results

Research questions

• RQ1 Can a sequence-based PPI prediction method achieve state-of-the-art accuracy while scaling to the entire human interactome?
• RQ2 How does SPRINT perform on the most challenging PPI prediction scenarios (C2 and C3 types), where no protein in the test pair appears in the training data?
• RQ3 What is the computational cost of whole-interactome PPI prediction using SPRINT compared to existing tools like Ding et al. and Martin et al.?
• RQ4 Can the use of spaced seeds and similarity propagation significantly improve prediction accuracy while maintaining low memory usage?
• RQ5 Is SPRINT capable of completing full human interactome prediction in a practical timeframe (e.g., under 2 hours) on standard hardware?

Key findings

• SPRINT achieves the highest accuracy among all evaluated sequence-based methods on all seven reliable human PPI datasets, especially excelling on the difficult C2 and C3 test sets.
• SPRINT completes full human interactome prediction in 15 to 100 minutes on a 12-core machine, while the next-fastest competitor (Ding et al.) requires weeks and others require years.
• SPRINT uses less than 1 GB of memory, whereas Ding et al.’s method fails on large datasets due to excessive memory demands.
• The method’s score propagation mechanism, based on Equation (3), effectively transfers interaction signals from known PPIs to novel pairs via shared sequence motifs.
• SPRINT is the only tool capable of predicting the entire human interactome in a practical timeframe, transforming a previously intractable problem into a routine task.
• The top 1% of predicted PPIs from SPRINT are made publicly available at www.csd.uwo.ca/faculty/ilie/SPRINT/ for community use.