{
  "id": 227757,
  "title": "COMPASS: Component-Wise Inference of Shared and Gene-Specific Perturbation Response",
  "url": "https://urgent.news/2026/08/06/compass-component-wise-inference-of-shared-and-gene-specific",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-06T00:00:00.000Z",
  "source": {
    "name": "bioRxiv",
    "slug": "biorxiv",
    "url": "https://www.biorxiv.org/content/10.64898/2026.08.03.742643v1?rss=1"
  },
  "original_language": "en",
  "account": "The study explores the intricate response of a cell's transcriptome to genetic perturbations, a crucial objective in computational biology. Despite various models achieving comparable accuracy in predicting mean responses across several perturbations, they fail to differentiate between individual perturbations. In this investigation, 2,270 CRISPRi perturbations were examined across six cell lines. The findings reveal that these perturbations follow a uniform pattern, oscillating between responses aligned with the mean and those deviating from it. Remarkably, a perturbation's position along this spectrum remains consistent across cell lines, with a Kendall's W coefficient of 0.59. Furthermore, this consistency is linked to STRING protein-interaction embeddings, demonstrating a predictive relationship (R^2=0.35). To encapsulate this organization, researchers have developed COMPASS, an interpretable linear model capable of segregating perturbation responses into shared and gene-specific elements. The shared-response component, a cell-line-wide response adjusted by a perturbation-specific coefficient, exhibits remarkable consistency across cell lines. This coefficient, strongly conserved, significantly contributes to the model's predictive power. The residual gene-specific component, though moderately conserved, captures pathway-level programs. When compared to scGPT, CPA, GEARS, GenePert, and State, COMPASS demonstrates superior performance in response accuracy, with a de-biased Pearson delta of 0.34.",
  "summary": "Predicting how a genetic perturbation reshapes a cell's transcriptome is a central goal of computational biology. Previous studies report that the mean response across training perturbations rivals specialized models on standard accuracy metrics, even though it cannot distinguish which perturbation occurred. Across 2,270 CRISPRi perturbations measured in each of six cell lines, we show that this…",
  "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."
}