{"ok":true,"nota":{"id":"aa694ac2519795aa","slug":"un-modelo-de-ia-reconstruye-la-forma-exacta-de-moleculas-biologicas-aa694a","titulo":"Un modelo de IA reconstruye la forma exacta de moléculas biológicas","resumen":"Crearon un sistema que convierte la forma tridimensional de proteínas, ADN, ARN y ligandos en una secuencia de símbolos, y que luego la reconstruye con alta precisión. Con ese sistema entrenaron un modelo que predice estructuras moleculares.","fuente":"biorxiv","fuenteNombre":"bioRxiv · preprints","color":"#be2628","autor":"Yan, J., Xie, D., Wang, Z., Xu, M., Zhang, Z.","url":"https://www.biorxiv.org/content/10.64898/2026.11.03.763913v1?rss=1","imagen":"/hoy/portada/aa694ac2519795aa","tema":"ciencia","subtemas":["estructura de proteinas","inteligencia artificial","biologia molecular","modelos predictivos"],"entidades":["CASP14","CASP15","Qwen3.5-9B","bioRxiv"],"formato":"dato","entretenimiento":false,"pubTs":"2026-11-04T00:00:00.000Z","indexTs":"2026-11-04T22:00:25.625Z","hash":"b368aa1c1dcea3502012b3ce476401717646d85c7705a77f6fad4a648b8e49cf","tsa":{"time":"Nov  4 22:00:41 2026 GMT","provider":"freetsa.org","sobre":"sello"},"stats":{"views":0,"likes":0,"clicks":0,"dwell":0},"tipo":"normal","registro":"hallazgo","porQueImporta":"Saber cómo se pliega una proteína o cómo se une a un fármaco define muchas investigaciones médicas. Este método traduce esa forma a un lenguaje que una IA puede manejar, parecido a cómo un texto se vuelve palabras. Para la ciencia básica y el desarrollo de medicamentos, eso reduce tiempos y costos. Los resultados se midieron en competencias internacionales, lo que da una base sólida, aunque siguen siendo predicciones que deben verificarse.","quienFirma":"Preprint en bioRxiv (sin institución principal identificada en el resumen)","confianza":"preprint","consulta":null},"sello":{"hash":"b368aa1c1dcea3502012b3ce476401717646d85c7705a77f6fad4a648b8e49cf","sello":"e629c2f562ea581bd87ac5ff712ba018c5eb32a4a94c19a8683bcf144b15f2de","beacon":{"drand":{"cadena":"8990e7a9aaed2ffed73dbd7092123d6f289930540d7651336225dc172e51b2ce","ronda":6613286,"aleatorio":"37d982263b1d863d3e73e8620b4d70d9d89b5524dd3510618f4f9548f0a32773","tiempo":"2026-11-04T22:00:00.000Z"},"nist":{"cadena":2,"pulso":2013577,"salida":"EE9469DC678F94C8E45728DDE99E7BA3500EC2481F5EA3B231150D35D58A60E297785CA9E7165F42698DB604B62391E7C5702183B1BA3EE13F4E2465594E4A68","tiempo":"2026-11-04T21:59:00.000Z"},"noAntes":"2026-11-04T22:00:00.000Z"},"verificacion":{"drand":true,"nist":true},"tsa":{"time":"Nov  4 22:00:41 2026 GMT","provider":"freetsa.org","sobre":"sello"},"original":{"titulo":"All-Atom Structure Tokenizer: AtomLM-9b for Biomolecular Structure Prediction","descripcion":"We develop an All-atom Structure Tokenizer that compresses biomolecular geometry into discrete structural tokens and achieves high-accuracy all-atom reconstruction. A hierarchical geometric encoder and finite scalar quantization map chemical units to a shared vocabulary of 8,192 tokens, while a joint coordinate de-tokenizer reconstructs backbone, side-chain, nucleic-acid, and ligand heavy atoms. The tokenizer achieves all-atom lDDT scores of 0.832 and 0.835 on CASP14 and CASP15, respectively. Building on this tokenizer, we train AtomLM-9b, an autoregressive language model based on Qwen3.5-9B that predicts structural tokens from molecular identities and connectivity and decodes them into comp","pubDate":"2026-11-04","url":"https://www.biorxiv.org/content/10.64898/2026.11.03.763913v1?rss=1"},"indexado":"2026-11-04T22:00:25.625Z"}}