{
  "id": 18500895,
  "title": "Predicting Transcriptional Responses to Unseen Compounds through Context Modeling and Multi-View Transfer",
  "url": "https://urgent.news/2026/11/06/predicting-transcriptional-responses-to-unseen-compounds-through",
  "topic": "science",
  "section": "Science",
  "published": "2026-11-06T00:00:00.000Z",
  "source": {
    "name": "bioRxiv",
    "slug": "biorxiv",
    "url": "https://www.biorxiv.org/content/10.64898/2026.11.05.764505v1?rss=1"
  },
  "original_language": "en",
  "account": "The paper titled \"Predicting Transcriptional Responses to Unseen Compounds through Context Modeling and Multi-View Transfer\" introduces a novel framework called UnDetect. Existing models struggle to predict drug-induced transcriptional responses when the compound is unknown, with accuracy dropping by just 0.012 Pearson correlation on a drug-blind L1000 benchmark. UnDetect addresses this issue by separating contextual variation from compound effects, enabling accurate prediction of responses to transcriptionally unprofiled compounds. The framework integrates chemical structure, annotated targets, mechanisms of action, and Cell Painting morphology into a normalized multi-view kernel, accommodating incomplete compound information. UnDetect transfers residual responses within and across profiled cell lines, allowing pharmacological and phenotypic relationships to inform predictions beyond chemical similarity. In a drug-blind benchmark, UnDetect outperformed a network that ignores the compound in every fold, achieving a 0.117 score compared to 0.092 for the strongest published model. On DeepCE s benchmark, UnDetect reached a 0.602 Pearson correlation, surpassing the reported 0.491 for the strongest published model. Transferring UnDetect across cell lines improved 28 out of 28 LINCS L1000 evaluations and 47 and 49 out of 50 Tahoe-100M cell-line evaluations at higher doses. The results demonstrate that separating context from compound effects and integrating complementary compound information enhances transcriptional prediction and expands the utility of perturbation atlases for virtual screening.",
  "summary": "Predicting drug-induced transcriptional responses could support in silico screening of new compounds, yet existing models barely use the compound: on a drug-blind L1000 benchmark, removing the drug input from seven published models changes accuracy by at most 0.012 Pearson correlation. They capture shared cellular and experimental variation while overlooking the compound-specific signals that…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}