Independent replication of 1-gene perturbation design: cumulative hits after 5 rounds of 128 genes
Replay of five CRISPR screens by authors independent of the agent, comparing the agent, the agent given randomly permuted feedback, and classical sequential designs.
Overview
Replay of five CRISPR screens by authors independent of the agent, comparing the agent, the agent given randomly permuted feedback, and classical sequential designs.
Consult the linked sources for architecture or protocol details. Missing evidence is not evidence of a missing capability.
55 recorded evaluations, 55 metric rows. A comparison chart has not yet been validated for these results. The table retains the individual findings and their sources.
Results
Results are available, but no reviewed comparison panel is linked in this release.
All evaluations
55 evaluations · 55 results. Different protocols are not a single leaderboard.
Filter evaluations
Applied filters: All linked evaluations
Source checking is not independent reproduction. Release 2026-10-10-7fcc3e48a123.
Methods and evaluation design
Procedure, tasks and evaluated configurations
Recorded evaluations
Each evaluation records what was tested and under which conditions.
- BDA (Llama-3.1-8B backbone) on Carnevale (Gupta et al. 2025)
- BDA (Llama-3.1-8B backbone) on Sanchez (Gupta et al. 2025)
- BDA (Llama-3.1-8B backbone) on Sanchez Down (Gupta et al. 2025)
- BDA (Llama-3.1-8B backbone) on IFNG (Gupta et al. 2025)
- BDA (Llama-3.1-8B backbone) on IL2 (Gupta et al. 2025)
- BDA (Qwen-2-7B backbone) on Carnevale (Gupta et al. 2025)
- BDA (Qwen-2-7B backbone) on Sanchez (Gupta et al. 2025)
- BDA (Qwen-2-7B backbone) on Sanchez Down (Gupta et al. 2025)
- BDA (Qwen-2-7B backbone) on IFNG (Gupta et al. 2025)
- BDA (Qwen-2-7B backbone) on IL2 (Gupta et al. 2025)
- BDA-Rand (Claude 3.5 Sonnet backbone) on Carnevale (Gupta et al. 2025)
- BDA-Rand (Claude 3.5 Sonnet backbone) on Sanchez (Gupta et al. 2025)
Baseline coverage
Reference methods help show what a model adds beyond simple controls. We track a null control and a conventional method for each protocol.
0 of 2 active baseline roles have published Rewire measurements in this release. Measurements on a selected protocol do not establish coverage of an entire suite.
No execution recipe linked to this protocol. Recipe availability does not establish a completed evaluation.
- External evaluations
- 50
- paper compilation
- 5
Literature evidence is not a Rewire measurement. Executed but unpublished runs and private review status are not included.
Null control
Proposed control: requires review
No-change prediction under matched control conditions
Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.
This is a suggested selection rule, not a validated method or a measured score.
Conventional reference
Proposed control: requires review
Training-only mean-effect or linear prediction
Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.
This is a suggested selection rule, not a validated method or a measured score.
Protocol coverage CSV (gzip) · Model evaluation matrix (gzip) · Source table (gzip) · Release and checksums (gzip)
Coverage is derived from release 2026-10-10-7fcc3e48a123. Source citations describe the original records; they do not validate an unreviewed baseline proposal. No results have been generated by this audit.
Run instructions
No runnable recipe has been reviewed for this protocol. Dataset access, model requirements, licences and compute requirements must be checked against its sources before execution.
Strengths, limitations and unresolved questions
Evidence
Source checking verifies the cited claim or transcription. It does not establish independent reproduction.
Evidence table
Inspect claims, sources and review details
Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.
One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.
0 evidence rows matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|
No evidence rows match these filters. Choose another scope or clear the search.
Sources and history
Release 2026-10-10-7fcc3e48a123 · Record review: source checked
1 source records and release history
- LLMs for Bayesian Optimization in Scientific Domains: Are We There Yet? · Original source · Findings of the Association for Computational Linguistics: EMNLP 2025, pages 15482-15510
Technical metadata and extraction receipts
Stable ID: tgtval-20261009-protocol-gupta2025-cumulative-hits-round5
- areas
- cells-tissues
- contexts
- research
- protocol
- At each of 5 rounds a method selects 128 genes; the measured response of each is revealed. Score: cumulative number of the screen's hits selected by the end of round 5, averaged over 5 runs. The agent is also run with randomly permuted outcomes (BDA-Rand) to test whether it uses the feedback at all; classical designs use the embeddings of the same language model as the agent's backbone.
- version
- Gupta et al. 2025 (Findings of EMNLP 2025) sections 4-6; Tables 1 and 2
- metric
- true-positive-count
- metric direction
- higher
- unit
- count
- unit detail
- Genes that are hits in the screen
- metric definition
- Cumulative count of the screen's hits selected across 5 rounds of 128 genes, averaged over 5 runs. Non-hits are genes the screen assayed that are not in its hit set.
- control
- BDA-Rand is the same agent with the mapping between genes and outcomes randomly permuted, at two levels: random measurement values, and random hit or not-hit feedback.
- limitations
- Offline replay of five of the same screens; untested genes are absent from the candidate pool rather than negative.; Cumulative hits is a count, so values are comparable only within a screen.; The Claude 3.5 Sonnet 'Reported Numbers' row is converted from Roohani et al. 2025, not re-run here; the source's own replication of that setting is lower on three of the four screens with a reported value (interleukin-2, interferon-gamma, tau) and higher on one (Carnevale).; The source does not print the hit threshold; it prints its own ground-truth hit counts per screen.
- source locator
- Sections 4.1, 5 and 6; Tables 1 and 2; appendix B.1.1
Related records
- uses data: Schmidt et al. 2022 screen, interleukin-2 production in primary human T cells
- uses data: Schmidt et al. 2022 screen, interferon-gamma production in primary human T cells
- uses data: Carnevale et al. 2022 screen, T-cell resistance to tumour-microenvironment inhibitory signals
- uses data: Sanchez et al. 2021 screen, endogenous tau protein level in neurons
- uses data: Sanchez et al. 2021 screen, decreased endogenous tau protein level in neurons
- subject: denominator: tgtval-20261009-protocol-gupta2025-cumulative-hits-round5
- subject: control_result: tgtval-20261009-protocol-gupta2025-cumulative-hits-round5
- subject: cross_source_consistency: tgtval-20261009-protocol-gupta2025-cumulative-hits-round5
- assessment: BDA (Llama-3.1-8B backbone) on Carnevale (Gupta et al. 2025)
- assessment: BDA (Llama-3.1-8B backbone) on Sanchez (Gupta et al. 2025)
- assessment: BDA (Llama-3.1-8B backbone) on Sanchez Down (Gupta et al. 2025)
- assessment: BDA (Llama-3.1-8B backbone) on IFNG (Gupta et al. 2025)
- assessment: BDA (Llama-3.1-8B backbone) on IL2 (Gupta et al. 2025)
- assessment: BDA (Qwen-2-7B backbone) on Carnevale (Gupta et al. 2025)
- assessment: BDA (Qwen-2-7B backbone) on Sanchez (Gupta et al. 2025)
- assessment: BDA (Qwen-2-7B backbone) on Sanchez Down (Gupta et al. 2025)
- assessment: BDA (Qwen-2-7B backbone) on IFNG (Gupta et al. 2025)
- assessment: BDA (Qwen-2-7B backbone) on IL2 (Gupta et al. 2025)
- assessment: BDA-Rand (Claude 3.5 Sonnet backbone) on Carnevale (Gupta et al. 2025)
- assessment: BDA-Rand (Claude 3.5 Sonnet backbone) on Sanchez (Gupta et al. 2025)
- assessment: BDA-Rand (Claude 3.5 Sonnet backbone) on Sanchez Down (Gupta et al. 2025)
- assessment: BDA-Rand (Claude 3.5 Sonnet backbone) on IFNG (Gupta et al. 2025)
- assessment: BDA-Rand (Claude 3.5 Sonnet backbone) on IL2 (Gupta et al. 2025)
- assessment: BDA-Rand (Llama-3.1-8B backbone) on Carnevale (Gupta et al. 2025)
- assessment: BDA-Rand (Llama-3.1-8B backbone) on Sanchez (Gupta et al. 2025)
- assessment: BDA-Rand (Llama-3.1-8B backbone) on Sanchez Down (Gupta et al. 2025)
- assessment: BDA-Rand (Llama-3.1-8B backbone) on IFNG (Gupta et al. 2025)
- assessment: BDA-Rand (Llama-3.1-8B backbone) on IL2 (Gupta et al. 2025)
- assessment: BDA-Rand (Qwen-2-7B backbone) on Carnevale (Gupta et al. 2025)
- assessment: BDA-Rand (Qwen-2-7B backbone) on Sanchez (Gupta et al. 2025)
- assessment: BDA-Rand (Qwen-2-7B backbone) on Sanchez Down (Gupta et al. 2025)
- assessment: BDA-Rand (Qwen-2-7B backbone) on IFNG (Gupta et al. 2025)
- assessment: BDA-Rand (Qwen-2-7B backbone) on IL2 (Gupta et al. 2025)
- assessment: BDA (Replicated) (Claude 3.5 Sonnet backbone) on Carnevale (Gupta et al. 2025)
- assessment: BDA (Replicated) (Claude 3.5 Sonnet backbone) on Sanchez (Gupta et al. 2025)
- assessment: BDA (Replicated) (Claude 3.5 Sonnet backbone) on Sanchez Down (Gupta et al. 2025)
- assessment: BDA (Replicated) (Claude 3.5 Sonnet backbone) on IFNG (Gupta et al. 2025)
- assessment: BDA (Replicated) (Claude 3.5 Sonnet backbone) on IL2 (Gupta et al. 2025)
- assessment: BDA (Reported Numbers) (Claude 3.5 Sonnet backbone) on Carnevale (Gupta et al. 2025)
- assessment: BDA (Reported Numbers) (Claude 3.5 Sonnet backbone) on Sanchez (Gupta et al. 2025)
- assessment: BDA (Reported Numbers) (Claude 3.5 Sonnet backbone) on Sanchez Down (Gupta et al. 2025)
- assessment: BDA (Reported Numbers) (Claude 3.5 Sonnet backbone) on IFNG (Gupta et al. 2025)
- assessment: BDA (Reported Numbers) (Claude 3.5 Sonnet backbone) on IL2 (Gupta et al. 2025)
- assessment: GP (Llama-3.1-8B backbone) on Carnevale (Gupta et al. 2025)
- assessment: GP (Llama-3.1-8B backbone) on Sanchez (Gupta et al. 2025)
- assessment: GP (Llama-3.1-8B backbone) on Sanchez Down (Gupta et al. 2025)
- assessment: GP (Llama-3.1-8B backbone) on IFNG (Gupta et al. 2025)
- assessment: GP (Llama-3.1-8B backbone) on IL2 (Gupta et al. 2025)
- assessment: GP (Qwen-2-7B backbone) on Carnevale (Gupta et al. 2025)
- assessment: GP (Qwen-2-7B backbone) on Sanchez (Gupta et al. 2025)
- assessment: GP (Qwen-2-7B backbone) on Sanchez Down (Gupta et al. 2025)
- assessment: GP (Qwen-2-7B backbone) on IFNG (Gupta et al. 2025)
- assessment: GP (Qwen-2-7B backbone) on IL2 (Gupta et al. 2025)
- assessment: Linear UCB (Llama-3.1-8B backbone) on Carnevale (Gupta et al. 2025)
- assessment: Linear UCB (Llama-3.1-8B backbone) on Sanchez (Gupta et al. 2025)
- assessment: Linear UCB (Llama-3.1-8B backbone) on Sanchez Down (Gupta et al. 2025)
- assessment: Linear UCB (Llama-3.1-8B backbone) on IFNG (Gupta et al. 2025)
- assessment: Linear UCB (Llama-3.1-8B backbone) on IL2 (Gupta et al. 2025)
- assessment: Linear UCB (Qwen-2-7B backbone) on Carnevale (Gupta et al. 2025)
- assessment: Linear UCB (Qwen-2-7B backbone) on Sanchez (Gupta et al. 2025)
- assessment: Linear UCB (Qwen-2-7B backbone) on Sanchez Down (Gupta et al. 2025)
- assessment: Linear UCB (Qwen-2-7B backbone) on IFNG (Gupta et al. 2025)
- assessment: Linear UCB (Qwen-2-7B backbone) on IL2 (Gupta et al. 2025)
- assessed by: Select therapeutic targets for validation