The new system, called TriCCA-PPI, short for Tri-modal Chained Cross-Attention Protein-Protein Interaction prediction, was developed by Yubao Liu, Haiyue Jiang, Chenhao Li, Benrui Wang, and Yinfei Dai. Its central insight is that a protein’s identity and behavior can be described from three fundamentally different angles: the evolutionary signature written into its amino acid sequence, the three-dimensional shape it folds into, and the functional annotations that biologists have accumulated about it over decades of research. Most existing computational tools use only one or two of these views. The Changchun team argues that this incomplete coverage is a major reason why current predictors generalize poorly when confronted with unfamiliar proteins.

The first of the three modalities is evolutionary sequence information, captured using ESM-2, a large protein language model trained on billions of natural sequences. Models like ESM-2 learn, in effect, the grammar of protein evolution, encoding which positions in a sequence tolerate change and which are conserved because they are critical for function. The second modality comes from ESM-IF1, an inverse folding model that works from protein structure, capturing the geometry of how a chain of amino acids packs into a three-dimensional fold. The third draws on the Gene Ontology, a controlled vocabulary of gene and protein functions, converted into numerical embeddings through a technique called GO-anc2vec, which represents each protein’s functional annotations together with those of its ancestors in the ontology hierarchy.

What makes TriCCA-PPI architecturally distinctive is not just that it uses all three modalities, but how it combines them. The framework employs a fully decoupled feature extraction pipeline, meaning each of the three embedding types is generated by an independent module. This modular design has a practical consequence: as better protein language models or structure predictors become available, each extractor can be swapped out and upgraded independently without rebuilding the entire system. In a field where the underlying models are improving at a breakneck pace, that kind of replaceability is more than an engineering nicety. It also means the framework can compensate for the incomplete biological characterization that plagues single- or dual-modal approaches.

The fusion itself is handled by a chained pairwise cross-attention module, a mechanism borrowed and adapted from modern deep learning. Rather than simply concatenating the three embeddings, or averaging them, the model performs three rounds of progressive alignment. In each round, one modality attends to another, learning which features of the sequence representation are relevant to the structural representation, which structural features relate to functional annotations, and so on. This layered process is designed to capture deep complementary relationships among signals that are, biologically speaking, orthogonal: a protein’s evolutionary history, its physical shape, and its annotated role in the cell each constrain the others, but none fully determines the rest. Shallow fusion strategies, the authors contend, miss precisely these cross-cutting dependencies.

Once the multimodal protein representations are built, TriCCA-PPI turns to the topology of the interaction network itself. It employs a Graph Isomorphism Network, or GIN, arranged in a global-local dual-channel configuration and enhanced with Jumping Knowledge aggregation. In plain terms, the model looks at each protein both in its immediate neighborhood of interaction partners and in the broader structure of the whole network, and it combines information gathered at multiple depths of message passing. This matters because interaction networks are not random graphs; they have hubs, clusters, and long-range organization that carry biological meaning. A predictor that ignores this topology throws away information that a graph-aware model can exploit.

Another practical challenge the researchers tackled is class imbalance. In real PPI datasets, some interaction categories are heavily overrepresented while others are rare, and standard training procedures tend to bias models toward the common categories. TriCCA-PPI applies an asymmetric loss function, a technique that penalizes errors on the minority classes more heavily and down-weights the contribution of easy, abundant examples. The result is a model that does not simply sacrifice the rare and biologically interesting interaction types on the altar of overall accuracy.

The evaluation was thorough by the standards of the field. The team tested TriCCA-PPI on two widely used benchmark datasets, SHS27K and SHS148K, which are human subsets of the STRING database of known and predicted interactions. Crucially, they used three different data-splitting strategies: Random, Breadth-First Search, and Depth-First Search. The Random split distributes proteins freely between training and test sets, while the BFS and DFS splits are deliberately harder, constructing test sets that are topologically separated from the training data and thus better proxies for real-world generalization to unfamiliar proteins. Across all three settings, TriCCA-PPI outperformed existing state-of-the-art methods, with the most substantial gains appearing exactly where previous models were weakest: on low-homology proteins, which share little sequence similarity with anything in the training data, and on rare PPI categories.

Ablation experiments, in which components of the model are systematically removed to measure their individual contributions, confirmed the design logic. Removing any one of the three modalities degraded performance, validating the claim that sequence, structure, and function each carry non-redundant information. The experiments also showed that chained cross-attention outperforms shallow fusion alternatives, supporting the argument that progressive, layered alignment of modalities is what unlocks the complementary relationships between them.

The implications reach well beyond a leaderboard. Accurate classification of PPI categories is critical for elucidating intracellular signaling pathways and disease mechanisms, and reliable computational prediction can substantially reduce the cost of high-throughput wet-lab screening. A framework that generalizes better to low-homology proteins is particularly valuable in the era of rapid genome sequencing, where newly discovered proteins from poorly studied organisms often have no close relatives in existing databases. And because the architecture is modular, interpretable in the sense that each modality’s contribution can be assessed, and extensible to future feature extractors, the authors position TriCCA-PPI not as a single fixed model but as a practical foundation for large-scale, multi-category interaction prediction. The research, published open access on 3 September 2026 and funded by the Natural Science Foundation Program of Jilin Province, arrives as biology increasingly becomes a data science, and it offers a template for how the flood of sequence, structure, and functional information now accumulating in public databases can be fused into something genuinely more predictive than the sum of its parts.

Subject of Research: Multimodal deep learning prediction of multi-category protein-protein interactions

Article Title: TriCCA‑PPI: chained cross‑attention based multimodal fusion for multi‑category protein‑protein interaction prediction

Article References: Liu, Y., Jiang, H., Li, C., Wang, B., & Dai, Y. (2026). TriCCA‑PPI: chained cross‑attention based multimodal fusion for multi‑category protein‑protein interaction prediction. BMC Bioinformatics. https://doi.org/10.1186/s12859-026-06615-9

Image Credits: AI Generated

DOI: 10.1186/s12859-026-06615-9

Keywords: protein-protein interaction, multimodal fusion, cross-attention, deep learning, graph neural network, ESM-2, protein structure, Gene Ontology, bioinformatics, STRING database, class imbalance, computational biology

The trouble is that most state-of-the-art prediction methods come with a hidden price tag. To squeeze out ever-better accuracy, recent models have grown dependent on a sprawling ecosystem of external data: pre-trained protein language models trained on billions of evolutionary sequences, three-dimensional structures fetched from repositories such as the Protein Data Bank, gene ontology annotations, tissue expression profiles, and more. Each additional data source can help, but it also creates fragility. If a laboratory works with a poorly studied organism whose proteins lack rich annotations, or if access to a heavyweight pre-trained model is impractical, the entire prediction pipeline can stumble. A second, less obvious problem concerns scale. Protein-protein interaction networks are graphs, and the relationships within them exist at multiple levels simultaneously, from the local neighborhood of a single protein to long-range dependencies that stretch across the entire network. Capturing those multi-scale patterns typically demands computationally expensive graph transformers, whose attention mechanisms scale quadratically with network size and quickly become prohibitive.

A research team led by Yan Kang and Hu Yuan of Yunnan University, together with Xinchao Wang of Southeast University and Shengzhe Yao of Yunnan University, set out to tackle both limitations head-on. Writing in the journal Applied Intelligence, they introduce NMPPI, a lightweight framework that predicts protein-protein interactions using only the intrinsic information already present in the PPI network itself, namely protein sequences and the topology of the interaction graph. No external databases, no auxiliary annotations, no giant pre-trained language models. The work, published as volume 56, article 459 of Applied Intelligence, demonstrates that careful architectural design can match or beat far heavier systems while consuming a fraction of the training time and memory. The datasets underpinning the study derive from the widely used STRING database of protein-protein associations, and the authors state that the code will be publicly released on GitHub upon publication.

The first pillar of NMPPI is an unsupervised feature extraction strategy built on Node2vec, a random-walk-based algorithm originally developed for learning representations of nodes in arbitrary networks. Rather than treating the PPI graph as a static collection of labeled points, Node2vec simulates biased random walks that wander from protein to protein along interaction edges, producing sequences of node visits that behave like sentences in a language. A skip-gram style embedding procedure then learns vector representations in which proteins that occupy similar structural roles, or that frequently co-occur within short walking distance, end up close together in the embedding space. The elegance of this approach is that it extracts meaningful structural information purely from the order in which nodes appear in the generated sequences, requiring no supervision and no external knowledge. In effect, the network topology itself becomes the training corpus, and the resulting embeddings encode a compact summary of each protein’s relational position within the interactome.

Sequence information complements these structural embeddings. Each protein’s amino acid sequence carries the biochemical fingerprint that determines which binding partners it can physically engage, and NMPPI incorporates sequential features alongside the Node2vec-derived structural vectors to form a unified representation of every node. This dual-channel design, pairing what a protein is with where it sits in the network, allows the model to reason about interactions from two independent angles. It also directly answers the question the authors pose about feature extraction when external data is unavailable: by relying exclusively on sequences and graph structure, NMPPI remains fully functional even for organisms or datasets where nothing else can be obtained. That property matters enormously for labs studying neglected pathogens, rare species, or newly sequenced genomes, where the ecosystem of annotations that powers mainstream predictors simply does not exist yet.

The second pillar, and the technical heart of the paper, is a supervised learning model that fuses a Graph Isomorphism Network with a novel Graph-Mamba module. Graph Isomorphism Networks, often abbreviated GIN, are among the most expressive message-passing architectures available for graph learning, capable in principle of distinguishing graph structures that fool simpler convolutional variants. They excel at aggregating local neighborhood information, building up each node’s representation layer by layer. But message passing alone has a well-known blind spot: information travels only one hop per layer, so capturing dependencies between proteins separated by many edges requires deep stacks of layers, which invite over-smoothing and vanishing gradients. Long-range dependencies, precisely the multi-scale relationships the team wanted to capture, have historically been expensive to model in graphs.

The Graph-Mamba module addresses that bottleneck with a state space model, a class of sequence model that has recently surged in popularity as an efficient alternative to transformers. Where transformer attention computes pairwise interactions between every pair of tokens at quadratic cost, state space models compress historical information into a fixed-size hidden state that is updated recursively, yielding linear computational complexity in sequence length. Applied to the PPI network, Graph-Mamba treats node representations as a sequence and selectively filters which information to retain or discard as it sweeps through, capturing inter-node dependencies across the entire graph without the combinatorial blow-up of attention. The result is a model that perceives both the fine-grained local structure handled by the GIN and the sweeping global context handled by the Mamba-style module, achieving genuine multi-scale relational learning at a computational budget that scales gently with network size. The authors report that this combination lets NMPPI extract long-range dependencies that purely local architectures miss, while sidestepping the memory wall that plagues graph transformers.

The empirical case for the framework rests on extensive experiments across two real PPI datasets. According to the paper, NMPPI outperforms state-of-the-art methods on both benchmarks, and the advantages become even more striking when the model is pushed outside its comfort zone. In cross-dataset scenarios, where a model trained on one interaction network is evaluated on another, NMPPI showed superior generalization performance, suggesting that its reliance on intrinsic sequence and structural features produces representations that transfer rather than overfit to dataset-specific quirks. The efficiency numbers are equally eye-catching: NMPPI required only 27 percent of the training time and 36 percent of the parameter size compared with advanced competing methods. For research groups without access to large GPU clusters, those reductions can mean the difference between running a meaningful experiment overnight versus not running it at all, and they hint at a broader trend in machine learning for biology where leaner, purpose-built architectures are beginning to challenge the assumption that bigger is always better.

Beyond the headline results, the study sits at a fascinating intersection of several research currents. Protein language models such as ESM-2 and ProGen have dominated recent headlines for their ability to predict mutation effects and infer atomic-level structure at evolutionary scale, and graph neural networks have become the default tool for interaction prediction, with prior efforts like hierarchical graph learning frameworks and attention-based hybrids pushing accuracy steadily upward. Meanwhile, the Mamba family of state space models has been spreading rapidly from language modeling into vision and, increasingly, into biology, with early efforts such as Protein-Mamba exploring biological applications. NMPPI’s contribution is to show that these ingredients can be combined in a way that deliberately minimizes external dependencies, a design philosophy that could prove especially valuable as the field grapples with the uneven distribution of high-quality biological data across the tree of life.

The practical implications stretch from basic science to drug discovery. More complete and accurate PPI maps help researchers identify previously unknown components of signaling pathways, prioritize candidate drug targets whose disruption would selectively cripple disease-relevant protein complexes, and interpret the growing flood of genomic data pouring out of sequencing pipelines. A predictor that works without external annotations lowers the barrier to entry for scientists working on understudied organisms, from emerging pathogens to agriculturally important crops, and its modest computational footprint makes large-scale screening feasible on ordinary hardware. The authors acknowledge no conflicts of interest, and the work was supported in part by the National Natural Science Foundation of China and a Yunnan Provincial Major Science and Technology Project. With the code slated for public release, the broader community will soon be able to test whether this lean, topology-first approach becomes a new baseline for interaction prediction, one that proves the most important clues about how proteins connect were hiding in the network all along.

Subject of Research: Deep learning prediction of protein-protein interactions using graph neural networks and state space models

Article Title: NMPPI: multi-scale relationship structure with effective pre-training for protein-protein interactions

Article References: Kang, Y., Yuan, H., Wang, X., & Yao, S. (2026). NMPPI: multi-scale relationship structure with effective pre-training for protein-protein interactions. Applied Intelligence, 56(15), Article 459. https://doi.org/10.1007/s10489-026-07339-2

Image Credits: AI Generated

DOI: 10.1007/s10489-026-07339-2

Keywords: protein-protein interactions, graph neural networks, Graph Isomorphism Network, Graph-Mamba, state space models, Node2vec, pre-training, computational biology, STRING database, multi-scale learning, protein sequences, machine learning