The study, led by Jorge F. Beltrán and colleagues, presents a multi-scale benchmark of ESM2 models for binary classification of viral entry proteins. The researchers compared three versions of the model containing approximately 650 million, 3 billion, and 15 billion parameters. These systems were not retrained on the viral-protein dataset. Instead, the investigators used “frozen embeddings”—fixed numerical descriptions generated by ESM2 for each protein sequence—and supplied those representations to conventional machine-learning classifiers. This approach separates the task of learning general protein patterns from the task of distinguishing entry proteins from other viral proteins.

Protein language models are inspired by the way large language models process written text. In a protein sequence, amino acids function like an alphabet, while recurring sequence patterns reflect structural elements, interaction sites, cellular targeting signals, and evolutionary constraints. During training, models such as ESM2 learn to predict missing or masked amino acids from their surrounding context. Although the model is not explicitly told that a sequence belongs to a fusion protein or receptor-binding protein, its internal representation can encode relationships associated with folding, topology, and molecular function. The resulting embedding is a high-dimensional numerical profile that can be analyzed by a separate classifier.

To create the benchmark, the researchers assembled 1,092 reviewed viral proteins from UniProt/Swiss-Prot. The dataset was balanced between viral entry proteins and non-entry proteins, reducing the risk that a classifier could appear accurate simply by favoring the more common class. To limit the influence of closely related sequences, the proteins were clustered with CD-HIT at 40 percent sequence identity. The resulting data were divided into a training set of 873 proteins and an independent test set of 219 proteins. This sequence-redundancy control is important because random splits can produce deceptively high scores when nearly identical proteins occur on both sides of the evaluation.

Five supervised algorithms were tested on the ESM2 embeddings: a support vector machine with a radial basis function kernel, random forest, XGBoost, LightGBM, and a multilayer perceptron. These classifiers differ in how they model decision boundaries. The radial-basis-function support vector machine can separate classes in a transformed feature space, while tree-based methods identify combinations of informative variables through ensembles of decision trees. The multilayer perceptron, by contrast, learns nonlinear transformations through artificial neural-network layers. Comparing several algorithms allowed the researchers to assess whether performance depended primarily on the protein representation or on the choice of downstream classifier.

Performance increased as the ESM2 model grew larger, although the improvement became modest beyond the intermediate scale. The strongest result came from the 15-billion-parameter ESM2 model combined with the radial-basis-function support vector machine. On the independent test set, this configuration achieved 88.6 percent accuracy, a Matthews correlation coefficient of 0.772, and a receiver operating characteristic area under the curve of 0.956. The Matthews correlation coefficient is particularly useful for binary classification because it incorporates true and false positives and negatives into a single measure and remains informative when class distributions are uneven. The ROC-AUC value indicates strong discrimination across a range of classification thresholds.

The 3-billion-parameter model produced nearly the same outcome when paired with the same classifier, reaching 88.1 percent accuracy, an MCC of 0.765, and a ROC-AUC of 0.953. This small difference suggests that most of the useful information for this task was captured before the largest model scale. In practical terms, the 3B model may offer a more attractive balance between accuracy, memory requirements, and computational cost. The findings also illustrate an important feature of protein language models: larger networks can improve biological representation, but their gains may diminish when the dataset is relatively small or when the classification problem is already well represented by an intermediate model.

The ESM2 systems also outperformed traditional sequence-based descriptors. The study evaluated amino-acid composition, dipeptide composition, and pseudo-amino-acid composition, which summarize proteins using frequencies of individual residues, adjacent residue pairs, or combinations of composition and sequence-order information. Such descriptors are computationally efficient and remain useful in many bioinformatics applications, but they discard much of the contextual information present in a full sequence. The best descriptor-based baseline reached an MCC of 0.537 and a ROC-AUC of 0.846 on the independent test set, considerably below the strongest ESM2 configurations. The comparison supports the idea that contextual embeddings can capture signals that simple frequency-based features overlook.

The authors further explored how the method behaved when evaluated across taxonomic boundaries. In a taxid-disjoint analysis using the 650-million-parameter ESM2 model and the support vector machine, the mean MCC was 0.734 and the mean ROC-AUC was 0.929. This experiment was designed to test whether the classifier could retain predictive ability when proteins associated with particular taxonomic identifiers were separated between training and evaluation. The result is encouraging for surveillance applications, where newly encountered viruses may not be represented in the training data. However, the researchers emphasize that the analysis does not establish performance on completely unseen viral families. Taxonomic separation at the identifier level can still leave evolutionary relationships between training and test proteins.

The study’s practical outcome is ViralEntryPred, a web application intended for rapid screening of protein sequences for possible viral entry function. Such a tool could help prioritize candidates for laboratory testing, structural modeling, receptor-binding studies, or vaccine research. It is not a replacement for experimental validation: a computational prediction cannot by itself demonstrate receptor usage, membrane-fusion activity, cellular tropism, or pathogenicity. The results are also bounded by the quality and scope of the curated dataset, the definition of an entry protein, and the possibility that sequence patterns associated with known viral groups may not generalize to deeply divergent organisms. Even with these limitations, the benchmark demonstrates how frozen protein-language-model embeddings can extend functional annotation into regions of viral sequence space where conventional similarity searches provide little guidance. As genomic surveillance expands, intermediate-scale models such as ESM2 3B could become useful components of a broader pipeline for identifying molecular features that shape the earliest and most decisive step of infection: viral entry.

Subject of Research: Viral entry protein prediction using protein language model embeddings and machine learning

Article Title: Protein language models for viral entry protein prediction: a multi-scale ESM2 benchmark

Article References: Beltrán, J. F., Belén, L. H., Lugo, A., et al. “Protein language models for viral entry protein prediction: a multi-scale ESM2 benchmark.” BMC Bioinformatics (2026).

Image Credits: AI Generated

DOI: 10.1186/s12859-026-06622-w

Keywords: Protein language models, ESM2, viral entry proteins, membrane fusion, machine learning, protein function prediction, embeddings

Recent advancements have paved the way for innovative approaches to tackle this issue. A new deep learning model, ImmunoStruct, emerges as a promising solution. This model harnesses the power of multimodal data to provide a more comprehensive prediction of multi-allele class I peptide–MHC immunogenicity. By synthesizing information from amino acid sequences, structural configurations, and biochemical properties, ImmunoStruct aims to transcend the limitations of existing predictive methodologies.

The design and implementation of ImmunoStruct were driven by a vast multimodal dataset comprising 26,049 peptide–MHC interactions. This extensive dataset serves as a foundational pillar upon which the model was built, allowing it to learn intricate patterns and relationships that would remain concealed in traditional approaches. The integration of different modalities—sequence, structural, and biochemical—significantly enhances the model’s ability to predict immunogenicity, demonstrating superior performance over existing paradigms.

One of the standout features of ImmunoStruct is its emphasis on both performance and interpretability. In the context of therapeutic development, the interpretability of predictive models is crucial. ImmunoStruct not only predicts which epitopes are likely to be immunogenic, but it also provides insights into why certain peptides are recognized by the immune system better than others. This dual capability can accelerate the design and testing of new vaccines, supporting researchers in making informed decisions throughout the therapeutic development process.

The utility of ImmunoStruct extends beyond academic interest; it has significant implications for real-world applications, particularly in response to urgent health crises. An application of the model was evaluated using a dataset of SARS-CoV-2 epitopes, where its predictions exhibited strong alignment with in vitro assay results. This alignment not only underscores the model’s predictive accuracy but also enhances confidence in its application for rapid vaccine development against emerging infectious diseases.

Further augmenting its relevance, ImmunoStruct has shown promising results in the context of cancer. In cases where peptide–MHC interactions are critical for immune recognition of tumor neoepitopes, the model has demonstrated its capability to predict survival outcomes for patients based on peptide–MHC interactions. This predictive power could prove invaluable in personalizing cancer treatment, guiding clinicians in selecting the most effective therapeutic strategies based on individual patient profiles.

Delving into the technical underpinnings of the ImmunoStruct architecture, the model leverages equivariant graph processing techniques. By employing graph-based representations, it can effectively capture the relationships between amino acids within a peptide and their corresponding positions in the MHC. This innovation allows the model to retain important spatial information, a crucial factor in understanding the immunogenic potential of different peptide configurations.

Moreover, the equilibrium achieved within the model through its multimodal data integration contributes to its robustness. Analyzing diverse forms of data allows ImmunoStruct to build a more nuanced view of epitope presentation and recognition. This holistic perspective is essential, as it not only aids in identifying strong immunogenic candidates but also helps in visualizing and interpreting the underlying biological mechanisms.

The development of ImmunoStruct has not emerged in a vacuum. The growing attention to deep learning methodologies in the biological sciences signifies a shift in how researchers approach data analysis and predictive modeling. By harnessing artificial intelligence, ImmunoStruct embodies the potential of these techniques, offering a glimpse into the future of immunotherapy and vaccine development, where machine learning tools play a central role.

As the scientific community continues to grapple with the challenges posed by infectious diseases and cancer, models like ImmunoStruct are becoming increasingly vital. The ability to predict immunogenicity with higher accuracy not only accelerates the vaccine development process but also enhances the potential for successful therapeutic outcomes. As ImmunoStruct gains traction, its application could extend to various other pathogens and cancers, dramatically transforming immunotherapy landscape.

In summary, ImmunoStruct represents a significant advancement in the field of immunology. By seamlessly integrating sequence, structural, and biochemical data through innovative deep learning techniques, it provides a powerful tool for predicting immunogenicity. The implications of this research extend far beyond academia, holding promise for enhancing vaccine development and personalizing cancer therapies. As future studies build upon this foundation, we may very well witness a transformation in how researchers approach the identification and development of therapeutics that could save countless lives.

Efforts to further enhance the capabilities of ImmunoStruct are underway. Researchers across the globe are keenly interested in expanding the dataset to cover more peptide–MHC combinations and refine the model’s accuracy. Additionally, collaborations between computational biologists and immunologists will be instrumental in translating these predictive insights into real-world vaccine formulations. Thus, the journey towards fully harnessing the potential of ImmunoStruct is only just beginning.

With its impressive capabilities and the promise of future refinements, ImmunoStruct stands at the frontier of immunogenicity prediction. As scientific exploration continues to unveil the complexities of the immune response, this deep learning model may very well be pivotal in ushering in a new era of personalized medicine and effective therapeutic interventions for infectious diseases and cancer.

Subject of Research: Immunogenicity prediction of peptide-MHC interactions

Article Title: ImmunoStruct enables multimodal deep learning for immunogenicity prediction

Article References:

Givechian, K.B., Rocha, J.F., Liu, C. et al. ImmunoStruct enables multimodal deep learning for immunogenicity prediction.
Nat Mach Intell (2025). https://doi.org/10.1038/s42256-025-01163-y

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s42256-025-01163-y

Keywords: Epitope-based vaccines, immunogenicity prediction, deep learning, peptide-MHC interactions, SARS-CoV-2, cancer neoepitopes, multimodal data integration.