{"ok":true,"nota":{"id":"24f0500b09e1acb8","slug":"benchmarks-que-miden-una-cosa-distinta-a-la-que-dicen-medir-24f050","titulo":"Benchmarks que miden una cosa distinta a la que dicen medir","resumen":"Investigadores crearon un sistema para evaluar cuáles intervenciones en células llevan a un estado deseado. Encontraron que una alta precisión en un tipo de prueba no garantiza buenos resultados en otra.","fuente":"biorxiv","fuenteNombre":"bioRxiv · preprints","color":"#be2628","autor":"Li, J., Huang, L. F., Song, W., Li, Y., Jakub, J. W., Qin, H., Luo, Y., Tao, C.","url":"https://www.biorxiv.org/content/10.64898/2026.10.19.760271v1?rss=1","imagen":"/hoy/portada/24f0500b09e1acb8","tema":"ciencia","subtemas":["inteligencia artificial","biología computacional","benchmarks","genómica","evaluación de modelos"],"entidades":["InverseBench","FSTN","K562"],"formato":"dato","entretenimiento":false,"pubTs":"2026-10-23T00:00:00.000Z","indexTs":"2026-10-23T18:12:31.737Z","hash":"06940f711becf088106e452645eecf4621908ca176de47aed9a4614ac020ce8f","tsa":{"time":"Oct 23 18:13:14 2026 GMT","provider":"freetsa.org","sobre":"sello"},"stats":{"views":0,"likes":0,"clicks":0,"dwell":0},"tipo":"normal","registro":"hallazgo","porQueImporta":"Muchos modelos de inteligencia artificial en biología se presentan con cifras impresionantes. Un ejemplo concreto: el mismo método acierta 0,959 en una medida y apenas 0,445 en crecimiento celular real. Para el público, es una señal de que un número alto no alcanza para confiar en una herramienta científica. Conviene preguntar siempre qué se midió exactamente.","quienFirma":"Preprint sin institución principal identificada","confianza":"preprint","consulta":null},"sello":{"hash":"06940f711becf088106e452645eecf4621908ca176de47aed9a4614ac020ce8f","sello":"6a811f4638bcbced87e5d8859ef55f4d523117795e947652874f766407ccdc6f","beacon":{"drand":{"cadena":"8990e7a9aaed2ffed73dbd7092123d6f289930540d7651336225dc172e51b2ce","ronda":6578272,"aleatorio":"e504aca3fcd7243fddbec8b88f1c51dc96cb1f827ccda23c013f3b302f3c69ab","tiempo":"2026-10-23T18:13:00.000Z"},"nist":{"cadena":2,"pulso":1996329,"salida":"880EE1B503674E57D0E81D7CEC0FFDBB3A3BB5CD5A62E05BB575BBCB9E2DB29F611799CDCF1B120D62E8DF3A39AE0D1B9935A0A198FC708C5974107E51CFF0C4","tiempo":"2026-10-23T18:12:00.000Z"},"noAntes":"2026-10-23T18:13:00.000Z"},"verificacion":{"drand":true,"nist":true},"tsa":{"time":"Oct 23 18:13:14 2026 GMT","provider":"freetsa.org","sobre":"sello"},"original":{"titulo":"Information contracts determine what inverse perturbation benchmarks measure","descripcion":"Inverse perturbation benchmarks rank interventions toward a desired cellular state, but high accuracy can answer a different question from biological prioritization. We present InverseBench, which makes candidate information and endpoint construction explicit, and FSTN, a linear pipeline evaluated under it. With identical inputs, candidates and validation, nested ridge reaches Spearman 0.959 for a transcriptional principal-component proxy but 0.445 for independently measured K562 growth fitness across 7,354 genes. Controlled response-entry availability across seven tasks gives heterogeneous gains, 0.442 for CD4 activation and 0.116 for external fitness. Yet predefined shortlist comparisons s","pubDate":"2026-10-23","url":"https://www.biorxiv.org/content/10.64898/2026.10.19.760271v1?rss=1"},"indexado":"2026-10-23T18:12:31.737Z"}}