---
format: "aidr-story-markdown/v1"
id: "9b1d61365688c400976131984dd91b599f0a9241635ec062c7127bb20d03d2f2"
canonical_url: "https://aidr.today/9b1d6136?lang=en"
title: "Do Perturbation Predictors Learn Gene Interactions? Diagnosing Extrapolation in Agentic Screening"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-15T04:00:00.000Z"
category: "Research"
topics: ["inference"]
source_urls: ["https://arxiv.org/abs/2610.18880"]
summary: "Across eight prediction methods and three Perturb-seq datasets, specialized neural predictors did not consistently outperform a simple additive baseline, and predictions with larger deviations from the additive baseline tended to have larger errors, while additive disagreement remained informative across compositional settings."
---

# Do Perturbation Predictors Learn Gene Interactions? Diagnosing Extrapolation in Agentic Screening

> [Open the canonical story](<https://aidr.today/9b1d6136?lang=en>)

**Published:** 2026-10-15T04:00:00.000Z
**Category:** Research
**Topics:** inference

## Summary

Across eight prediction methods and three Perturb\-seq datasets, specialized neural predictors did not consistently outperform a simple additive baseline, and predictions with larger deviations from the additive baseline tended to have larger errors, while additive disagreement remained informative across compositional settings\.

## Sources

- [Story source](<https://arxiv.org/abs/2610.18880>)