Parallel Protein Community Detection in Large-scale PPI Networks Based on Multi-source Learning

Jianguo Chen, Kenli Li, Kashif Bilal, Ahmed A. Metwally, Keqin Li, Philip S. YuIEEE/ACM 2018 · DOI · OpenReview

Abstract

Protein interactions constitute the fundamental building block of almost every life activity. Identifying protein communities from Protein-Protein Interaction (PPI) networks is essential to understand the principles of cellular organization and explore the causes of various diseases. It is critical to integrate multiple data resources to identify reliable protein communities that have biological significance and improve the performance of community detection methods for large-scale PPI networks. In this paper, we propose a Multi-source Learning based Protein Community Detection (MLPCD) algorithm by integrating Gene Expression Data (GED) and a parallel solution of MLPCD using cloud computing technology. GED under different conditions is integrated with the original PPI network to reconstruct a Weighted-PPI network. To flexibly identify protein communities of different scales, we define community modularity and functional cohesion measurements and detect protein communities from WPPI. In addition, we compare the detected communities with known protein complexes and evaluate the function enrichment of protein functional modules using Gene Ontology annotations. We implement a parallel version of MLPCD on the Apache Spark platform to enhance the performance of the algorithm. Extensive experimental results indicate the superiority and notable advantages of the MLPCD algorithm over the relevant algorithms in terms of accuracy and performance.

Introduction

Proteins rarely act in isolation; they form complex "communities" to perform vital cellular functions. Traditionally, researchers have identified these communities using Protein-Protein Interaction (PPI) networks—graphs where nodes represent proteins and edges represent physical interactions. However, traditional topological methods face two major hurdles: biological noise and scalability. PPI networks are often static and noisy, failing to account for the fact that proteins are regulated by dynamic biological processes. Furthermore, as high-throughput sequencing generates massive datasets, serial algorithms become prohibitively slow for networks containing millions of interactions.

The MLPCD Framework

The MLPCD algorithm addresses these challenges by integrating topological data with biological context. The process operates in four primary stages, as illustrated in the pipeline overview below:

Overview of the MLPCD algorithm pipeline, from data integration to community analysis.
  1. Weighted-PPI (WPPI) Construction: Instead of treating all interactions as equal, the algorithm uses Gene Expression Data (GED) to calculate the Pearson correlation coefficient between proteins. High "co-expression" weights indicate interactions more likely to occur in real biological contexts.
  2. Community Initialization: MLPCD identifies "hub proteins" (nodes with high weighted degrees) to serve as initial seeds, providing a more stable starting point than random node assignment.
  3. Modularity Optimization: The algorithm employs an agglomerative approach to merge nodes into communities by optimizing a custom Community Modularity ($QC_k$) metric.
  4. Functional Cohesion Refinement: To ensure biological relevance, the authors define Functional Cohesion ($FC_k$), which balances connectivity with interaction intensity.

Parallel Implementation on Apache Spark

To handle large-scale datasets, the entire pipeline is implemented on the Apache Spark platform using the GraphX library. The implementation utilizes a "vertex-cut" partitioning method to distribute the graph across a cluster. The parallel construction of the WPPI network is managed through a series of Resilient Distributed Dataset (RDD) transformations, ensuring that the heavy computational load of calculating correlation coefficients and edge weights is distributed efficiently.

The RDD-based parallel workflow for constructing the Weighted-PPI network on Apache Spark.

Experimental Evaluation

The authors evaluated MLPCD using 10 species datasets, including S. cerevisiae (yeast) and H. sapiens (human), sourced from databases like DIP and STRING. Validation was performed against known protein complexes in the MIPS database and Gene Ontology (GO) annotations.

  • Accuracy: In yeast tests, MLPCD identified 405 matched complexes, significantly outperforming SPICi (342) and Label Propagation (187).
  • Biological Significance: Communities detected by MLPCD showed lower P-values in GO enrichment analysis, indicating they correspond to actual biological functions like DNA replication or metabolic processes.
  • Performance: On a mouse dataset with 6.3 million interactions, MLPCD completed in approximately 42 seconds. The parallel implementation achieved a speedup of up to 18.2x with a 20-node cluster, demonstrating near-linear scalability.

Conclusion

MLPCD successfully filters biological noise by integrating multi-source data and overcomes scalability barriers through parallel computing. While the current model significantly improves community detection, future work will focus on incorporating dynamic characteristics and additional data sources, such as protein domain information, to further refine the accuracy of detected functional modules.

Summary updated

Abstract

Protein interactions constitute the fundamental building block of almost every life activity. Identifying protein communities from Protein-Protein Interaction (PPI) networks is essential to understand the principles of cellular organization and explore the causes of various diseases. It is critical to integrate multiple data resources to identify reliable protein communities that have biological significance and improve the performance of community detection methods for large-scale PPI networks. In this paper, we propose a Multi-source Learning based Protein Community Detection (MLPCD) algorithm by integrating Gene Expression Data (GED) and a parallel solution of MLPCD using cloud computing technology. To effectively discover the biological functions of proteins that participating in different cellular processes, GED under different conditions is integrated with the original PPI network to reconstruct a Weighted-PPI (WPPI) network. To flexibly identify protein communities of different scales, we define community modularity and functional cohesion measurements and detect protein communities from WPPI using an agglomerative method. In addition, we respectively compare the detected communities with known protein complexes and evaluate the functional enrichment of protein function modules using Gene Ontology annotations. Moreover, we implement a parallel version of the MLPCD algorithm on the Apache Spark platform to enhance the performance of the algorithm for large-scale realistic PPI networks. Extensive experimental results indicate the superiority and notable advantages of the MLPCD algorithm over the relevant algorithms in terms of accuracy and performance.
BibTeX
@inproceedings{chen2018parallel,
  author = {Jianguo Chen and Kenli Li and Kashif Bilal and Ahmed A. Metwally and Keqin Li and Philip S. Yu},
  title = {{Parallel Protein Community Detection in Large-scale PPI Networks Based on Multi-source Learning}},
  booktitle = {IEEE/ACM},
  year = {2018},
  doi = {10.1109/tcbb.2018.2868088},
}
Figure