A protein’s function depends not just on its own structure and localization, but also on the interactions with its partners. Many proteins are therefore better described by a set of partner-dependent roles than by a single annotation. Yet most approaches to the functional interpretation of protein-protein interactions (PPIs) remain protein or set-centric. They rely on pre-existing annotations, and perform worst where knowledge is sparse. Here, we present MAPPIE (Map of Protein-Protein Interaction Embeddings), a method that treats each PPI, rather than each protein, as a unit of representation. From 199,137 human interactions spanning 15,503 proteins, we build a two-dimensional map of the human PPI landscape for functional discovery. Protein language model embeddings for two protein interaction partners are combined and compressed into a latent space, with model selection guided by domain-domain interactions used as a structural proxy for interaction similarity. The resulting geometry separates domain defined interaction classes, organizes disorder associated interactions spatially, and splits interactions involving the same protein by partner. A query PPI’s latent neighbourhood recovers its own annotated functions across molecular, complex, pathway, and biological processes. MAPPIE contributes most where existing functional evidence is weakest, outperforming interactome and sequence identity baselines for sparsely connected interactions. MAPPIE neighbours of query PPIs are enriched for partners in independent protein networks, recovering curated complex-level function even when subunits are spread across the map. Applied to a human dark interactome, MAPPIE assigns specific, experimentally supported functions to dark hub proteins.
A neural network-based pipeline that integrates amino acid sequences with structural features is developed and provides a modular prototype for follow-up, more extensive protein modeling, including larger proteins and sequence of variable sizes.
Carl David Jasper Causin, M. Fyta· APL Machine Learning· 0 citations
In computational structural biology, it is still very hard to accurately find protein-protein interaction (PPI) sites and estimate how strong the interaction would be. In this research, we provide an innovative two-stage deep learning framework that combines residue-level graph representation learning with protein-level regression to achieve a thorough modeling of protein interactions. Protein structures first encoded as residue graphs, with nodes that stand for amino acids and edges that show how close they are to each other in space. To find binding residues, a deep residual Graph Attention Network v2 (GATv2) uses multi-head attention, residual connections, and Jumping Knowledge aggregation to collect long-range relationships and structural information at different scales. Using residue-level predictions, protein embeddings are created and put together to provide paired representations that show how similar and different two interacting proteins are. After that, these representations utilized to train a regression model that can predict continuous interaction strength ratings. The proposed model tested using a huge human PPI dataset that has 2,242 complexes. The proposed model performs very well at the residue level, with an AUROC of 0.9625, an AUPRC of 0.9149, an F1-score of 0.8192, and an MCC of 0.7674. The protein-level regression model also does a great job of predicting, with an RMSE of 0.2806, an MAE of 0.1635, and a R2 of 0.6500 on the test set. It also has high correlation coefficients (Pearson = 0.8069, Spearman = 0.7500), which means that the predicted and true interaction strengths are very similar. In general, the proposed model is a single, scalable approach that connects predicting binding sites at the residue level with estimating interaction strength at the protein level. This gives us a better understanding of the structural processes that control PPI.
Oras A. Hussein, E. Al-Shamery· Journal of Intelligent Infor...· 0 citations
The functional classification of protein sequences remains a major bottleneck in biology. Although protein language model (PLM)-based approaches have substantially improved broad protein function prediction, most protein sequences still lack precise annotation at the level of specialized functions—the fine-grained molecular roles that define specificity within protein families. We present PLMView, an unsupervised framework for fine-grained protein function classification directly from sequence. PLMView reframes protein function inference as a relational problem: instead of embedding sequences in isolation, it positions them within a collaborative functional space defined by comparisons with PLM embeddings of anchor sequences, thereby capturing subtle sequence–function relationships. Without requiring labeled data, family-specific training, or PLM fine-tuning, PLMView accurately distinguishes specialized functions among homologous proteins and highlights residues likely to determine functional specificity. The method achieves high precision while remaining computationally efficient, classifying approximately 10,000 sequences with 1,000 anchors in under 40 minutes; compared with pooled-embedding approaches and, in challenging cases, Sequence Similarity Networks, PLMView provides finer and more biologically coherent functional resolution, while achieving more than 10-fold speed-up over SSN reconstruction on datasets of this scale. Applications to thioredoxins, visual opsins, and Tara Oceans environmental diatom cold-shock proteins show that PLMView can move from interpretable residue-level determinants in well-studied protein families to large-scale environmental functional discovery, linking molecular specialization to ecological distribution and transcriptional deployment across the global ocean.
Multiple sequence alignment (MSA) Pairformer is presented, a protein language model that builds on AlphaFold2/3's bidirectional refinement between sequence and pairwise residue representations to accurately model the evolution of protein-protein interactions, despite training exclusively on individual chains.
Yo Akiyama, Zhidian Zhang, Olivia Tang et al.· Cell· 2 citations
Accurate protein-protein interaction (PPI) prediction is central to functional genomics, disease mechanism discovery, and drug development. A difficult setting arises when candidate interactions include proteins that have no observed PPI edges during training, where models relying on network topology alone often lose useful context. This paper presents \method, a multimodal representation framework for cold-start PPI prediction. \method\ combines region-aware protein sequence encoding with four protein-centered biomedical knowledge graphs, including protein-drug, protein-disease, protein-miRNA, and protein-lncRNA associations. The sequence branch extracts contextual representations from structurally informed sequence regions, while graph attention encoders learn modality-specific protein embeddings from sparse biomedical associations. A bridge reconstruction objective regularizes graph learning by recovering shared protein-entity associations, and a pair-level gating module adaptively integrates sequence and graph evidence for each candidate protein pair. Experiments on two benchmark datasets under novel-old and novel-novel cold-start settings show that \method\ consistently outperforms competitive sequence, network, and knowledge-graph baselines across ACC, F1, AUC, AUPR, and MCC.
Mapping the protein interactome is fundamental to understanding disease mechanisms and facilitating therapeutic development. Although protein language models (PLMs) such as ESM-2 have advanced protein-protein interaction (PPI) prediction, their high-dimensional representations remain difficult to connect to verifiable biological signals. To address this limitation, we propose HybridStack-PPI, a gray-box framework that combines ESM-2 sequence representations with explicit physicochemical and motif-derived biological descriptors. The architecture uses motif-anchored local pooling global mean pooling, symmetric pair encoding, fold-internal feature selection, LightGBM branch learners, and an elastic-net logistic-regression stacking layer. We evaluated the method using a C3 cluster-based cross-validation protocol with a 40% sequence-identity clustering threshold and a Same-GO hard-negative setting in which negative candidates shared functional annotations with positive pairs. Under this setting, HybridStack-PPI reached a Human ROC-AUC of 73.65%, PR-AUC of 91.35%, MCC of 28.06%, and specificity of 75.61%. The results indicate a conservative operating point: compared to more recall-oriented baselines, the proposed stack trades lower recall and F1 for higher specificity, MCC, and ranking behavior under functionally similar negative samples. We further reported cross-species transfer, ablation, latency, SHAP-based descriptor attribution, and meta-learner coefficient analyses to clarify both the promise and limitations of biologically informed PPI prediction.
T. T. Nguyen, X. Mai, N. Nguyen· IEEE Access· 0 citations