Hard and Soft GMM-EM for Overlapping Protein Complex Detection Across Five Disease-Specific PPI Networks

Anif Hanifa Setianingrum; Wisnu Ananta Kusuma; Agus Buono; Karlisa Priandana

doi:10.15408/jti.v19i2.52476

Authors

Anif Hanifa Setianingrum Informatic Engineering, Faculty Science and Technology, University of Islamic Syarif Hidayatullah, Jakarta, Indonesia
Wisnu Ananta Kusuma Bioinformatics Study Program, Faculty of Mathematics and Natural Sciences, IPB University Indonesia
Agus Buono Doctoral Program in Computer Science, School of Data Science Mathematic and Informatics, IPB University, Bogor, Indonesia
Karlisa Priandana Doctoral Program in Computer Science, School of Data Science Mathematic and Informatics, IPB University, Bogor, Indonesia

DOI:

https://doi.org/10.15408/jti.v19i2.52476

Keywords:

Hard-Soft Clustering, Gaussian Mixture Model Expectation Maximization, Overlapping protein complexes, Diseases- specific PPI networks

Abstract

Detecting overlapping protein complexes in disease-specific protein–protein interaction (PPI) networks remains a significant challenge, as multifunctional proteins may participate in more than one biological module. This study compares conventional hard and probabilistic soft assignments within a common Gaussian Mixture Model–Expectation Maximization (GMM–EM) framework across five disease-specific PPI networks. Proteins were represented by seven topology-based centrality features, and the optimal number of components was determined using an automatically detected Bayesian Information Criterion (BIC) elbow. Hard assignment used the highest posterior probability, whereas soft assignment retained memberships at τ = 0.10, 0.20, or 0.30. Clusters were evaluated against CORUM using Jaccard, Precision, Recall, F1-score, and Overlap Coefficient. Performance was network-dependent: ALPS achieved the highest Jaccard (0.0852–0.0903), Precision (0.0934–0.0979), and F1-score (0.1555–0.1638); leukemia the highest Recall/Overlap (0.6375–0.6386); breast cancer the largest mean intersection (16.62–16.64); HCC showed intermediate performance; and lung cancer the lowest Jaccard/F1 but greatest membership uncertainty, with 16.1% retaining a second posterior probability ≥0.10. The soft assignment increased mean Recall in leukemia (0.4333–0.4601) and lung cancer (0.3799–0.4030). Aggregate F1 improved from 0.315 to 0.321 (Wilcoxon p=0.0038), and permutation testing showed a mean F1 gain of 0.018 (10,000 permutations, p=0.012). Soft assignment mainly preserves secondary membership information, with benefits depending on network structure. Limitations include static STRING networks, topology-only features, CORUM-based validation, unrecorded runtime, and no experimental verification.

References

[1] G. Liu, B. Liu, A. Li, X. Wang, J. Yu, and X. Zhou, “Identifying Protein Complexes With Clear Module Structure Using Pairwise Constraints in Protein Interaction Networks,” Front. Genet., vol. 12, no. August, pp. 1–18, 2021, doi: 10.3389/fgene.2021.664786.

[2] G. Tsitsiridis et al., “CORUM: the comprehensive resource of mammalian protein complexes-2022,” Nucleic Acids Res., vol. 51, no. D1, pp. D539–D545, 2023, doi: 10.1093/nar/gkac1015.

[3] F. Liang et al., “Recent advances and challenges in protein complex model accuracy estimation,” Comput. Struct. Biotechnol. J., vol. 23, no. January, pp. 1824–1832, 2024, doi: 10.1016/j.csbj.2024.04.049.

[4] N. Zhao, T. Wu, W. Wang, L. Zhang, and X. Gong, “Review and Comparative Analysis of Methods and Advancements in Predicting Protein Complex Structure,” Interdiscip. Sci. - Comput. Life Sci., vol. 16, no. 2, pp. 261–288, 2024, doi: 10.1007/s12539-024-00626-x.

[5] A. G. Gasic, A. Sarkar, and M. S. Cheung, “Understanding protein-complex assembly through grand canonical maximum entropy modeling,” Phys. Rev. Res., vol. 3, no. 3, 2021, doi: 10.1103/PhysRevResearch.3.033220.

[6] M. Mustafa et al., “Apoptosis: A Comprehensive Overview of Signaling Pathways, Morphological Changes, and Physiological Significance and Therapeutic Implications,” Cells, vol. 13, no. 22, pp. 1–29, 2024, doi: 10.3390/cells13221838.

[7] H. Oda, A. Annibaldi, D. L. Kastner, and I. Aksentijevich, “Genetic Regulation of Cell Death: Insights from Autoinflammatory Diseases,” Annu. Rev. Immunol., vol. 43, no. 1, pp. 313–342, 2025, doi: 10.1146/annurev-immunol-090222-105848.

[8] H. Mao, X. Zhao, and S. C. Sun, “NF-κB in inflammation and cancer,” Cell. Mol. Immunol., vol. 22, no. 8, pp. 811–839, 2025, doi: 10.1038/s41423-025-01310-w.

[9] Z. Zhang, J. Song, J. Tang, X. Xu, and F. Guo, “Detecting complexes from edge-weighted PPI networks via genes expression analysis,” BMC Syst. Biol., vol. 12, no. Suppl 4, 2018, doi: 10.1186/s12918-018-0565-y.

[10] J. Wang, J. Liang, W. Zheng, X. Zhao, and J. Mu, “Protein complex detection algorithm based on multiple topological characteristics in PPI networks,” Inf. Sci. (Ny)., vol. 489, pp. 78–92, 2019, doi: 10.1016/j.ins.2019.03.015.

[11] R. Wang, G. Liu, and C. Wang, “Identifying protein complexes based on an edge weight algorithm and core-attachment structure,” BMC Bioinformatics, vol. 20, no. 1, 2019, doi: 10.1186/s12859-019-3007-y.

[12] N. Zaki, H. Singh, and E. A. Mohamed, “Identifying Protein Complexes in Protein-Protein Interaction Data Using Graph Convolutional Network,” IEEE Access, vol. 9, pp. 123717–123726, 2021, doi: 10.1109/ACCESS.2021.3110845.

[13] E. Alvarez-Mamani, R. Dechant, C. A. Beltran-Castañón, and A. J. Ibáñez, “Graph embedding on mass spectrometry- and sequencing-based biomedical data,” BMC Bioinformatics, vol. 25, no. 1, pp. 1–19, 2024, doi: 10.1186/s12859-023-05612-6.

[14] J. Cui, S. Yang, L. Yi, Q. Xi, D. Yang, and Y. Zuo, “Recent advances in deep learning for protein-protein interaction: a review,” BioData Min., vol. 18, no. 1, 2025, doi: 10.1186/s13040-025-00457-6.

[15] L. Zhu, J. Lei, L. Klei, B. Devlin, and K. Roeder, “Semisoft clustering of single-cell data,” Proc. Natl. Acad. Sci. U. S. A., vol. 116, no. 2, pp. 466–471, 2019, doi: 10.1073/pnas.1817715116.

[16] M. Chimienti et al., “The use of an unsupervised learning approach for characterizing latent behaviors in accelerometer data,” Ecol. Evol., vol. 6, no. 3, pp. 727–741, 2016, doi: 10.1002/ece3.1914.

[17] H. Kim, S. W. Hyun, G. Hoogenboom, C. H. Porter, and K. S. Kim, “Fuzzy union to assess climate suitability of annual ryegrass (Lolium multiflorum), Alfalfa (Medicago sativa) and Sorghum (Sorghum bicolor),” Sci. Rep., vol. 8, no. 1, pp. 1–15, 2018, doi: 10.1038/s41598-018-28291-3.

[18] A. J. Enright, S. Van Dongen, and C. A. Ouzounis, “An efficient algorithm for large-scale detection of protein families,” 2002. doi: 10.1093/nar/30.7.1575.

[19] G. D. Bader and C. W. V. Hogue, “An automated method for finding molecular complexes in large protein interaction networks,” BMC Bioinformatics, vol. 4, 2003, doi: 10.1186/1471-2105-4-2.

[20] T. Nepusz, H. Yu, and A. Paccanaro, “Detecting overlapping protein complexes in protein-protein interaction networks,” Nat. Methods, vol. 9, no. 5, pp. 471–472, 2012, doi: 10.1038/nmeth.1938.

[21] B. Adamcsek, G. Palla, I. J. Farkas, I. Derényi, and T. Vicsek, “CFinder: Locating cliques and overlapping modules in biological networks,” Bioinformatics, vol. 22, no. 8, pp. 1021–1023, 2006, doi: 10.1093/bioinformatics/btl039.

[22] F. Consonni, E. Gambineri, and C. Favre, “ALPS, FAS, and beyond: from inborn errors of immunity to acquired immunodeficiencies,” Ann. Hematol., vol. 101, no. 3, pp. 469–484, 2022, doi: 10.1007/s00277-022-04761-7.

[23] G. S. Winkler, K. W. Mulder, V. J. Bardwell, E. Kalkhoven, and H. T. M. Timmers, “Human Ccr4-Not complex is a ligand-dependent repressor of nuclear receptor-mediated transcription,” EMBO J., vol. 25, no. 13, pp. 3089–3099, 2006, doi: 10.1038/sj.emboj.7601194.

[24] C. E. Hershberger et al., “HHS Public Access,” vol. 35, no. 4, pp. 1108–1120, 2021, doi: 10.1038/s41375-020-1002-y.Complex.

[25] N. Goossens, X. Sun, and Y. Hoshida, “Molecular classification of hepatocellular carcinoma: potential therapeutic implications,” Hepatic Oncol., vol. 2, no. 4, pp. 371–379, 2015, doi: 10.2217/hep.15.26.

[26] L. Peng, R. Gao, L. Han, T. Fu, and C. Bian, “Roles of β‑catenin protein in non‑small cell lung cancer (Review),” Oncol. Lett., vol. 30, no. 6, pp. 1–11, 2025, doi: 10.3892/ol.2025.15311.