Dr. Aarón Ramírez

A new RNA foundation model, published in Nature. Researchers introduce structRFM, an open, structure-guided RNA foundation model trained on 21 million RNA sequence–structure pairs. Unlike conventional RNA language models that primarily learn from sequences, structRFM incorporates base-pairing information directly into pre-training, enabling joint learning of RNA sequence and structure. Key highlights: • Structure-guided pre-training: A new masked language modeling strategy (SGMLM) captures RNA base-pairing relationships. • RNA structure prediction: Achieves state-of-the-art performance on secondary structure benchmarks. • 3D RNA folding: Its Zfold adaptation delivers competitive performance with AlphaFold 3 across established RNA structure benchmarks. • Broad generalization: Supports RNA structure prediction, splice-site prediction, IRES identification, and ncRNA classification. The complete dataset, model weights, and source code are openly available. From sequence-only learning to structure-aware foundation models, RNA AI is taking another step forward. Nature paper: doi.org/10.1038/s41586-026-1… #RNA #AIforBiology #FoundationModels #Nature

A virtual cell that predicts proteins, not just RNA. Published in Nature, ProteinTalks is trained on >38 million temporal protein-abundance measurements from drug perturbations. The key shift: from static expression snapshots to modeling how protein networks evolve over time — and using that to predict drug response, combinations, resistance and patient-specific outcomes. Virtual cells are starting to look less like benchmarks, and more like operational models of biology. nature.com/articles/s41586-0…