{"paper":{"title":"CellFluxRL: Biologically-Constrained Virtual Cell Modeling via Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Reinforcement learning post-training with biological rewards improves virtual cell generators to respect physical and biological rules.","cross_cats":["q-bio.QM"],"primary_cat":"cs.LG","authors_text":"Dongxia Wu, Elaine Sui, Emily B. Fox, Emma Lundberg, Serena Yeung-Levy, Shiye Su, Yuhui Zhang","submitted_at":"2026-03-23T09:33:18Z","abstract_excerpt":"Building virtual cells with generative models to simulate cellular behavior in silico is emerging as a promising paradigm for accelerating drug discovery. However, prior image-based generative approaches can produce implausible cell images that violate basic physical and biological constraints. To address this, we propose to post-train virtual cell models with reinforcement learning (RL), leveraging biologically meaningful evaluators as reward functions. We design seven rewards spanning three categories-biological function, structural validity, and morphological correctness-and optimize the st"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"CellFluxRL consistently improves over CellFlux across all rewards, with further performance boosts from test-time scaling, advancing beyond visually realistic generations towards biologically meaningful ones.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The seven reward functions accurately capture biologically meaningful constraints without introducing unintended biases or allowing the model to game the rewards while still violating real cellular physics.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CellFluxRL post-trains the CellFlux generative model with reinforcement learning driven by biologically meaningful reward functions, yielding virtual cell images that better satisfy physical and biological constraints than the base model.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Reinforcement learning post-training with biological rewards improves virtual cell generators to respect physical and biological rules.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"be02fedcba7bdd0c01dc9680bedeb8104529a9d947c899f806edcfc4690dc368"},"source":{"id":"2603.21743","kind":"arxiv","version":4},"verdict":{"id":"bb411121-dfa1-4da1-9cee-c48a8b514e64","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T01:01:16.403684Z","strongest_claim":"CellFluxRL consistently improves over CellFlux across all rewards, with further performance boosts from test-time scaling, advancing beyond visually realistic generations towards biologically meaningful ones.","one_line_summary":"CellFluxRL post-trains the CellFlux generative model with reinforcement learning driven by biologically meaningful reward functions, yielding virtual cell images that better satisfy physical and biological constraints than the base model.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The seven reward functions accurately capture biologically meaningful constraints without introducing unintended biases or allowing the model to game the rewards while still violating real cellular physics.","pith_extraction_headline":"Reinforcement learning post-training with biological rewards improves virtual cell generators to respect physical and biological rules."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2603.21743/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}