{
  "abstract": "Evaluating machine learning in scientific domains requires separating correct predictions from correct reasons under realistic distribution shifts. We introduce PertReason, a knowledge-grounded approach to evaluating and improving cell-state-conditioned reasoning about perturbation effects. Our framework comprises PertReasonQA, a benchmark that provides evaluation and supervision, and PertReasonLM, a reference large language model that demonstrates how such supervision supports reasoning-centric training. PertReasonQA tests whether models can generate mechanistically faithful explanations while remaining robust to complex shifts, such as new cells, unseen perturbations, cross-modal or combinatorial extrapolations. PertReasonQA combines single-cell genetic and chemical perturbation data across multiple cellular contexts with knowledge graphs, and dynamically conditions pathways on cell-specific basal states to avoid generic memorization. Evaluations of state-of-the-art models reveal systematic gaps between outcome prediction and mechanistic reasoning. As a reference probe of the benchmark, PertReasonLM is trained through supervised fine-tuning and reinforcement learning using mechanistic supervision from PertReasonQA. PertReasonLM improves predictive accuracy and reduces answer-reason disagreement by grounding rationales in context-specific pathways. Together, we provide a diagnostic framework for exposing and mitigating failures in faithful reasoning in data-rich scientific systems. We release our datasets to support further research at https://dongkwan-kim.github.io/PertReason/.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2607.18777",
  "author_total": 4,
  "authors": [
    {
      "name": "Dongkwan Kim"
    },
    {
      "name": "Yiming Gao"
    },
    {
      "name": "Yining Yang"
    },
    {
      "name": "Yang Shen"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/bydqs00m5gy4a.bib",
  "cited_by": 1,
  "doi": "10.48550/arxiv.2607.18777",
  "fulltext_url": null,
  "id": "bydqs00m5gy4a",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2607.18777",
  "language": null,
  "licence": "LicenseRef-arXiv-nonexclusive-1.0",
  "licence_label": "arXiv licence",
  "licence_source": "article",
  "licence_url": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.LG"
    ],
    [
      "arxiv",
      "q-bio.MN"
    ],
    [
      "openalex",
      "1312|Gene Regulatory Network Analysis"
    ]
  ],
  "openalex": "W7170089005",
  "pdf_url": "https://arxiv.org/pdf/2607.18777",
  "pmcid": null,
  "pmid": null,
  "published": "2026-11-11",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/bydqs00m5gy4a.ris",
  "sources": [
    {
      "first_seen": "2026-11-13T06:21:00Z",
      "native_id": "2607.18777",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2607.18777"
    }
  ],
  "subjects": [
    {
      "field": "biology",
      "origin": "arxiv:q-bio.MN",
      "subfield": "systems-synthetic"
    }
  ],
  "title": "PertReason: A Knowledge-Grounded Benchmark and Framework for Cell-State-Conditioned Mechanistic Reasoning of Perturbation Effects",
  "title_text": "PertReason: A Knowledge-Grounded Benchmark and Framework for Cell-State-Conditioned Mechanistic Reasoning of Perturbation Effects",
  "type": "preprint",
  "updated_at": "2026-11-13T06:21:00Z",
  "url": "https://research.draconitic.com/paper/bydqs00m5gy4a",
  "venue": "arXiv",
  "versions": []
}