Uploaded July 2026 | Updated November 2026, 3 days ago
Speaker:
María Rodríguez Martínez, Associate Professor of Biomedical Informatics & Data Science at Yale School of Medicine and a member of the Center for Systems and Engineering Immunology.
Abstract:
Deep learning has transformed protein science, enabling increasingly accurate structure prediction, protein language modeling, and generative protein design. Yet immune receptors remain a particularly difficult frontier: for antibodies and T cell receptors, accurate global folds are not sufficient. Binding specificity, cross-reactivity, and affinity are governed by flexible interfaces where current structure prediction models often have the lowest confidence.
In this talk, María discusses their work developing machine learning approaches to model B and T cell receptor recognition. She first introduces how molecular dynamics-informed generative models can capture TCR conformational ensembles beyond static structures, with implications for predicting peptide recognition and cross-reactivity. She then describes graph-based models of antibody–antigen affinity that integrate structural representations with protein language model embeddings to improve generalization across antibody families and antigens.
Finally, she presents ongoing work on TCR–peptide binding prediction, de novo design of pMHC binders, and fine-tuning protein language models to improve zero-shot ranking of immune receptor–ligand interactions.
Together, these examples illustrate a central theme: the next generation of immune receptor modeling will require moving beyond static structure prediction toward models that integrate sequence, structure, dynamics, and biophysical constraints at the binding interface.
Further information here:
bsc.es/research-and-development/research-seminars/sorswomeninbsc-machine-learning-driven-modeling-b-and-t-cells-receptors
Seminar within the Severo Ochoa Women Researchers Fellowship Programme.
Speaker:
María Rodríguez Martínez, Associate Professor of Biomedical Informatics & Data Science at Yale School of Medicine and a member of the Center for Systems and Engineering Immunology.
Abstract:
Deep learning has transformed protein science, enabling increasingly accurate structure prediction, protein language modeling, and generative protein design. Yet immune receptors remain a particularly difficult frontier: for antibodies and T cell receptors, accurate global folds are not sufficient. Binding specificity, cross-reactivity, and affinity are governed by flexible interfaces where current structure prediction models often have the lowest confidence.
In this talk, María discusses their work developing machine learning approaches to model B and T cell receptor recognition. She first introduces how molecular dynamics-informed generative models can capture TCR conformational ensembles beyond static structures, with implications for predicting peptide recognition and cross-reactivity. She then describes graph-based models of antibody–antigen affinity that integrate structural representations with protein language model embeddings to improve generalization across antibody families and antigens.
Finally, she presents ongoing work on TCR–peptide binding prediction, de novo design of pMHC binders, and fine-tuning protein language models to improve zero-shot ranking of immune receptor–ligand interactions.
Together, these examples illustrate a central theme: the next generation of immune receptor modeling will require moving beyond static structure prediction toward models that integrate sequence, structure, dynamics, and biophysical constraints at the binding interface.
Further information here:
bsc.es/research-and-development/research-seminars/sorswomeninbsc-machine-learning-driven-modeling-b-and-t-cells-receptors
Seminar within the Severo Ochoa Women Researchers Fellowship Programme.










