Leandro A. Bugnon

> AI x BIO

prof_pic.jpg

I develop machine learning methods to study the relationship between sequence, structure, and function in RNA and proteins. I am interested in how large sequence datasets can support learning when experimental data are scarce, and how to assess whether these models generalize to new molecules.

My work includes RNA structure prediction, protein function annotation, and representation learning that integrates sequence and structure. I also explore conditional generative models and active learning for molecular design and candidate selection.

I am an Associate Researcher at sinc(i)/UNL – CONICET, where I also serve as deputy director, and an Associate Professor at the National University of the Litoral and Austral University. I also collaborate with The Cell Company on applied bioinformatics projects.

When I’m not training neural networks I’m usually proofing dough — some baking algorithms to.

$ cat activity.log | tail -5
2026-11 [conference] 3rd IDPfun2 Conference @ XV CA2B2C, Buenos Aires · short talkShort talk at the 3rd IDPfun2 Conference, held alongside the XV meeting of the Argentine Association of Bioinformatics and Computational Biology, in the session on protein disorder, flexible linkers and computational tools. 2026-10 [scholarship] AISAR scholarship · AI safety for scientific agentsF. Peitti’s project under my supervision: did the agent do what it says it did? Evaluating and mitigating undeclared deviations in autonomous scientific agents. 2026-06 [preprint] Preprint: how should we cross-validate RNA folding models?A critical review of cross-validation strategies for RNA secondary structure prediction, and the homology leakage that makes results look better than they are. 2026-04 [visit] BioComputingUP @ Università di Padova (IDPfun2)Research visit to the BioComputingUP group within the IDPfun2 project: computational characterisation of linkers in intrinsically disordered proteins. 2026-04 [paper] ET-Pfam out in BioinformaticsEnsembles of transfer-learning classifiers that cut the protein family prediction error to 7%.
$ ▌