Evaluation

Summ. Emb. on the Virtual CRISPR Difficult Benchmark

Evidence concern: excluded from comparisons

Appendix A.2 describes the random baseline as predicting hit with probability equal to 'the overall proportion of actual hits in the specific benchmark dataset'. With one pooled rate (485 of 1,175 = 41.3%, Table 3) the expected PPV, sensitivity, F1 and FPR are all about 0.41 and specificity and NPV about 0.59. Table 8 prints PPV 0.47, sensitivity 0.46, F1 0.47, FPR 0.36, specificity 0.64 and NPV 0.63. With each screen's own hit rate (Table 3) the expected values are 0.47, 0.47, 0.47, 0.37, 0.63 and 0.63, which match. One draw at the pooled rate would need sensitivity about 2 standard errors above and FPR about 3 below expectation at once. The baseline therefore probably uses per-screen hit rates, which is information a screen designer would not have before the screen. The expected values were computed by the reviewer from Table 3 counts and are not stored as results. Treat the random row as a per-screen-prevalence classifier, not a matched random-selection comparator.Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Appendix A.2; Table 3; Table 1, row "Random Baseline"; Table 8, row "Random N/A N/A"

Published comparison; transcribed, not reproduced.

Evaluation results

1 evaluation · 8 results. Different protocols are not a single leaderboard.

Protocol: Virtual CRISPR Difficult Benchmark: hit prediction for complex phenotypes · Dataset: Virtual CRISPR Difficult Benchmark

Sorted by AUPRC (Difficult Benchmark) (higher is better). The best value in each column is highlighted. Decimals are rounded for display; each value links to the printed value and its source.

Tested configuration	AUPRC (Difficult Benchmark)	AUROC (Difficult Benchmark)	F1 (Difficult Benchmark)	False positive rate (Difficult Benchmark)	Negative predictive value (Difficult Benchmark)	Precision (Difficult Benchmark)	Recall (Difficult Benchmark)	Specificity (Difficult Benchmark)
Virtual CRISPR embedding classifier, GPT-4o summary embeddings (Song et al. 2025)Evaluation	0.69	0.72	0.67	0.26	0.67	0.71	0.64	0.74

All 8 result rows with coverage, uncertainty and sources

Filter evaluations

Tested model or methodProtocolDatasetEvaluation setupMetricEvidence origin

Applied filters: All linked evaluations

Exact evaluated configurations and original reported results
Tested configuration	Protocol and dataset	Finding	Evidence and details
Configuration: Virtual CRISPR embedding classifier, GPT-4o summary embeddings (Song et al. 2025)	Protocol: Virtual CRISPR Difficult Benchmark: hit prediction for complex phenotypes Dataset: Virtual CRISPR Difficult Benchmark	0.69 AUPRC (Difficult Benchmark) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix A.2 describes the random baseline as predicting hit with probability equal to 'the overall proportion of actual hits in the specific benchmark dataset'. With one pooled rate (485 of 1,175 = 41.3%, Table 3) the expected PPV, sensitivity, F1 and FPR are all about 0.41 and specificity and NPV about 0.59. Table 8 prints PPV 0.47, sensitivity 0.46, F1 0.47, FPR 0.36, specificity 0.64 and NPV 0.63. With each screen's own hit rate (Table 3) the expected values are 0.47, 0.47, 0.47, 0.37, 0.63 and 0.63, which match. One draw at the pooled rate would need sensitivity about 2 standard errors above and FPR about 3 below expectation at once. The baseline therefore probably uses per-screen hit rates, which is information a screen designer would not have before the screen. The expected values were computed by the reviewer from Table 3 counts and are not stored as results. Treat the random row as a per-screen-prevalence classifier, not a matched random-selection comparator.Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Appendix A.2; Table 3; Table 1, row "Random Baseline"; Table 8, row "Random N/A N/A" Methods, coverage and source Summ. Emb. on the Virtual CRISPR Difficult Benchmark pheno-20261022-protocol-song2025-difficult-hit-classification Aggregation: Not reported Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Table 2, row "Summ. Emb.", column "AUPRC"
Configuration: Virtual CRISPR embedding classifier, GPT-4o summary embeddings (Song et al. 2025)	Protocol: Virtual CRISPR Difficult Benchmark: hit prediction for complex phenotypes Dataset: Virtual CRISPR Difficult Benchmark	0.72 AUROC (Difficult Benchmark) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix A.2 describes the random baseline as predicting hit with probability equal to 'the overall proportion of actual hits in the specific benchmark dataset'. With one pooled rate (485 of 1,175 = 41.3%, Table 3) the expected PPV, sensitivity, F1 and FPR are all about 0.41 and specificity and NPV about 0.59. Table 8 prints PPV 0.47, sensitivity 0.46, F1 0.47, FPR 0.36, specificity 0.64 and NPV 0.63. With each screen's own hit rate (Table 3) the expected values are 0.47, 0.47, 0.47, 0.37, 0.63 and 0.63, which match. One draw at the pooled rate would need sensitivity about 2 standard errors above and FPR about 3 below expectation at once. The baseline therefore probably uses per-screen hit rates, which is information a screen designer would not have before the screen. The expected values were computed by the reviewer from Table 3 counts and are not stored as results. Treat the random row as a per-screen-prevalence classifier, not a matched random-selection comparator.Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Appendix A.2; Table 3; Table 1, row "Random Baseline"; Table 8, row "Random N/A N/A" Methods, coverage and source Summ. Emb. on the Virtual CRISPR Difficult Benchmark pheno-20261022-protocol-song2025-difficult-hit-classification Aggregation: Not reported Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Table 2, row "Summ. Emb.", column "AUROC"
Configuration: Virtual CRISPR embedding classifier, GPT-4o summary embeddings (Song et al. 2025)	Protocol: Virtual CRISPR Difficult Benchmark: hit prediction for complex phenotypes Dataset: Virtual CRISPR Difficult Benchmark	0.67 F1 (Difficult Benchmark) fraction · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix A.2 describes the random baseline as predicting hit with probability equal to 'the overall proportion of actual hits in the specific benchmark dataset'. With one pooled rate (485 of 1,175 = 41.3%, Table 3) the expected PPV, sensitivity, F1 and FPR are all about 0.41 and specificity and NPV about 0.59. Table 8 prints PPV 0.47, sensitivity 0.46, F1 0.47, FPR 0.36, specificity 0.64 and NPV 0.63. With each screen's own hit rate (Table 3) the expected values are 0.47, 0.47, 0.47, 0.37, 0.63 and 0.63, which match. One draw at the pooled rate would need sensitivity about 2 standard errors above and FPR about 3 below expectation at once. The baseline therefore probably uses per-screen hit rates, which is information a screen designer would not have before the screen. The expected values were computed by the reviewer from Table 3 counts and are not stored as results. Treat the random row as a per-screen-prevalence classifier, not a matched random-selection comparator.Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Appendix A.2; Table 3; Table 1, row "Random Baseline"; Table 8, row "Random N/A N/A" Methods, coverage and source Summ. Emb. on the Virtual CRISPR Difficult Benchmark pheno-20261022-protocol-song2025-difficult-hit-classification Aggregation: Not reported Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Table 2, row "Summ. Emb.", column "F1"
Configuration: Virtual CRISPR embedding classifier, GPT-4o summary embeddings (Song et al. 2025)	Protocol: Virtual CRISPR Difficult Benchmark: hit prediction for complex phenotypes Dataset: Virtual CRISPR Difficult Benchmark	0.26 False positive rate (Difficult Benchmark) fraction · lower Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix A.2 describes the random baseline as predicting hit with probability equal to 'the overall proportion of actual hits in the specific benchmark dataset'. With one pooled rate (485 of 1,175 = 41.3%, Table 3) the expected PPV, sensitivity, F1 and FPR are all about 0.41 and specificity and NPV about 0.59. Table 8 prints PPV 0.47, sensitivity 0.46, F1 0.47, FPR 0.36, specificity 0.64 and NPV 0.63. With each screen's own hit rate (Table 3) the expected values are 0.47, 0.47, 0.47, 0.37, 0.63 and 0.63, which match. One draw at the pooled rate would need sensitivity about 2 standard errors above and FPR about 3 below expectation at once. The baseline therefore probably uses per-screen hit rates, which is information a screen designer would not have before the screen. The expected values were computed by the reviewer from Table 3 counts and are not stored as results. Treat the random row as a per-screen-prevalence classifier, not a matched random-selection comparator.Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Appendix A.2; Table 3; Table 1, row "Random Baseline"; Table 8, row "Random N/A N/A" Methods, coverage and source Summ. Emb. on the Virtual CRISPR Difficult Benchmark pheno-20261022-protocol-song2025-difficult-hit-classification Aggregation: Not reported Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Table 2, row "Summ. Emb.", column "FPR"
Configuration: Virtual CRISPR embedding classifier, GPT-4o summary embeddings (Song et al. 2025)	Protocol: Virtual CRISPR Difficult Benchmark: hit prediction for complex phenotypes Dataset: Virtual CRISPR Difficult Benchmark	0.67 Negative predictive value (Difficult Benchmark) fraction · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix A.2 describes the random baseline as predicting hit with probability equal to 'the overall proportion of actual hits in the specific benchmark dataset'. With one pooled rate (485 of 1,175 = 41.3%, Table 3) the expected PPV, sensitivity, F1 and FPR are all about 0.41 and specificity and NPV about 0.59. Table 8 prints PPV 0.47, sensitivity 0.46, F1 0.47, FPR 0.36, specificity 0.64 and NPV 0.63. With each screen's own hit rate (Table 3) the expected values are 0.47, 0.47, 0.47, 0.37, 0.63 and 0.63, which match. One draw at the pooled rate would need sensitivity about 2 standard errors above and FPR about 3 below expectation at once. The baseline therefore probably uses per-screen hit rates, which is information a screen designer would not have before the screen. The expected values were computed by the reviewer from Table 3 counts and are not stored as results. Treat the random row as a per-screen-prevalence classifier, not a matched random-selection comparator.Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Appendix A.2; Table 3; Table 1, row "Random Baseline"; Table 8, row "Random N/A N/A" Methods, coverage and source Summ. Emb. on the Virtual CRISPR Difficult Benchmark pheno-20261022-protocol-song2025-difficult-hit-classification Aggregation: Not reported Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Table 2, row "Summ. Emb.", column "NPV"
Configuration: Virtual CRISPR embedding classifier, GPT-4o summary embeddings (Song et al. 2025)	Protocol: Virtual CRISPR Difficult Benchmark: hit prediction for complex phenotypes Dataset: Virtual CRISPR Difficult Benchmark	0.71 Precision (Difficult Benchmark) fraction · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix A.2 describes the random baseline as predicting hit with probability equal to 'the overall proportion of actual hits in the specific benchmark dataset'. With one pooled rate (485 of 1,175 = 41.3%, Table 3) the expected PPV, sensitivity, F1 and FPR are all about 0.41 and specificity and NPV about 0.59. Table 8 prints PPV 0.47, sensitivity 0.46, F1 0.47, FPR 0.36, specificity 0.64 and NPV 0.63. With each screen's own hit rate (Table 3) the expected values are 0.47, 0.47, 0.47, 0.37, 0.63 and 0.63, which match. One draw at the pooled rate would need sensitivity about 2 standard errors above and FPR about 3 below expectation at once. The baseline therefore probably uses per-screen hit rates, which is information a screen designer would not have before the screen. The expected values were computed by the reviewer from Table 3 counts and are not stored as results. Treat the random row as a per-screen-prevalence classifier, not a matched random-selection comparator.Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Appendix A.2; Table 3; Table 1, row "Random Baseline"; Table 8, row "Random N/A N/A" Methods, coverage and source Summ. Emb. on the Virtual CRISPR Difficult Benchmark pheno-20261022-protocol-song2025-difficult-hit-classification Aggregation: Not reported Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Table 2, row "Summ. Emb.", column "PPV"
Configuration: Virtual CRISPR embedding classifier, GPT-4o summary embeddings (Song et al. 2025)	Protocol: Virtual CRISPR Difficult Benchmark: hit prediction for complex phenotypes Dataset: Virtual CRISPR Difficult Benchmark	0.64 Recall (Difficult Benchmark) fraction · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix A.2 describes the random baseline as predicting hit with probability equal to 'the overall proportion of actual hits in the specific benchmark dataset'. With one pooled rate (485 of 1,175 = 41.3%, Table 3) the expected PPV, sensitivity, F1 and FPR are all about 0.41 and specificity and NPV about 0.59. Table 8 prints PPV 0.47, sensitivity 0.46, F1 0.47, FPR 0.36, specificity 0.64 and NPV 0.63. With each screen's own hit rate (Table 3) the expected values are 0.47, 0.47, 0.47, 0.37, 0.63 and 0.63, which match. One draw at the pooled rate would need sensitivity about 2 standard errors above and FPR about 3 below expectation at once. The baseline therefore probably uses per-screen hit rates, which is information a screen designer would not have before the screen. The expected values were computed by the reviewer from Table 3 counts and are not stored as results. Treat the random row as a per-screen-prevalence classifier, not a matched random-selection comparator.Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Appendix A.2; Table 3; Table 1, row "Random Baseline"; Table 8, row "Random N/A N/A" Methods, coverage and source Summ. Emb. on the Virtual CRISPR Difficult Benchmark pheno-20261022-protocol-song2025-difficult-hit-classification Aggregation: Not reported Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Table 2, row "Summ. Emb.", column "Sens."
Configuration: Virtual CRISPR embedding classifier, GPT-4o summary embeddings (Song et al. 2025)	Protocol: Virtual CRISPR Difficult Benchmark: hit prediction for complex phenotypes Dataset: Virtual CRISPR Difficult Benchmark	0.74 Specificity (Difficult Benchmark) fraction · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix A.2 describes the random baseline as predicting hit with probability equal to 'the overall proportion of actual hits in the specific benchmark dataset'. With one pooled rate (485 of 1,175 = 41.3%, Table 3) the expected PPV, sensitivity, F1 and FPR are all about 0.41 and specificity and NPV about 0.59. Table 8 prints PPV 0.47, sensitivity 0.46, F1 0.47, FPR 0.36, specificity 0.64 and NPV 0.63. With each screen's own hit rate (Table 3) the expected values are 0.47, 0.47, 0.47, 0.37, 0.63 and 0.63, which match. One draw at the pooled rate would need sensitivity about 2 standard errors above and FPR about 3 below expectation at once. The baseline therefore probably uses per-screen hit rates, which is information a screen designer would not have before the screen. The expected values were computed by the reviewer from Table 3 counts and are not stored as results. Treat the random row as a per-screen-prevalence classifier, not a matched random-selection comparator.Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Appendix A.2; Table 3; Table 1, row "Random Baseline"; Table 8, row "Random N/A N/A" Methods, coverage and source Summ. Emb. on the Virtual CRISPR Difficult Benchmark pheno-20261022-protocol-song2025-difficult-hit-classification Aggregation: Not reported Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Table 2, row "Summ. Emb.", column "Spec."

Source checking is not independent reproduction. Release 2026-10-23-5347695500ed.

Research readiness

0 of 4 readiness checks met. These checks assess whether the evidence supports a reproducible investigation; a source-checked score alone does not meet them.

Readiness checks, gaps and artifacts

Release 2026-10-23-5347695500ed · Evidence verified: Not verified

Evidence incomplete

Replay metrics

Exact outcomes, predictions, identifiers and evaluator are connected.

Missing or unresolved evidence

No verified artifact manifest is linked to this exact record.
File checksums match the recorded files: not yet verified
Predictions are matched to the right samples: not yet verified
Score meaning and direction are confirmed: not yet verified
Metrics are recomputed from the saved predictions: not yet verified

Verified: Not verified

Evidence incomplete

Investigate discrepancies

Replay evidence includes annotations and an assessment of dependence. Unknown independence permits descriptive analysis only.

Missing or unresolved evidence

No verified artifact manifest is linked to this exact record.
File checksums match the recorded files: not yet verified
Predictions are matched to the right samples: not yet verified
Score meaning and direction are confirmed: not yet verified
Metrics are recomputed from the saved predictions: not yet verified
Sample annotations are recorded: not yet verified
Dependence between samples is assessed: not yet verified

Verified: Not verified

Evidence incomplete

Run locally

A pinned recipe describes the inputs, environment and resource requirements.

Missing or unresolved evidence

No verified artifact manifest is linked to this exact record.
File checksums match the recorded files: not yet verified
Predictions are matched to the right samples: not yet verified
Score meaning and direction are confirmed: not yet verified
A pinned run recipe exists: not yet verified
Compute requirements are estimated: not yet verified

Verified: Not verified

Evidence incomplete

Validate independently

Separate data and exposure records support an independent test.

Missing or unresolved evidence

No verified artifact manifest is linked to this exact record.
File checksums match the recorded files: not yet verified
Predictions are matched to the right samples: not yet verified
Score meaning and direction are confirmed: not yet verified
Independent validation data exist: not yet verified
Overlap with training data is checked: not yet verified

Verified: Not verified

Readiness describes the evidence in this release. Availability on your computer is checked separately when an investigation runs. Existing data exposure can prevent independent validation even when files are available.

Artifacts and reproduction

No verified artifact manifest is connected to this record yet. The gaps above identify what is needed before analysis can begin.

Evaluation procedure

Virtual CRISPR Difficult Benchmark: hit prediction for complex phenotypes

Configuration: Virtual CRISPR embedding classifier, GPT-4o summary embeddings (Song et al. 2025)
Protocol: Virtual CRISPR Difficult Benchmark: hit prediction for complex phenotypes
Dataset: Virtual CRISPR Difficult Benchmark

origin: Author-reported evaluation
configuration: Primary source as retrieved 2026-10-22
protocol id: pheno-20261022-protocol-song2025-difficult-hit-classification
dataset version: Curated from PMID 39567689 and PMID 39385035
split: Fixed evaluation set of 1,814 queries; no data from BioGRID-ORCS
population: 1,814 gene-phenotype queries, 907 positive and 907 negative
inputs: Text of the perturbation method, gene symbol, cell line and phenotype, each embedded with text-embedding-3-large (raw text, or a GPT-4o summary for Summ. Emb.) and concatenated
metric implementation: AUROC and AUPRC from predicted probabilities; other metrics after binarising with Youden's J statistic
budget: Not reported
adaptation: Supervised MLP on LLM embeddings, trained on BioGRID-ORCS
aggregation: Single trained model; one evaluation

Metadata review: source checked. Unreported conditions prevent automatic comparisons.

Reproduction

Split: Fixed evaluation set of 1,814 queries; no data from BioGRID-ORCS
Adaptation: Supervised MLP on LLM embeddings, trained on BioGRID-ORCS
Scoring implementation: AUROC and AUPRC from predicted probabilities; other metrics after binarising with Youden's J statistic

No execution recipe has been verified for this exact configuration and evaluation. A benchmark's general instructions may use different inputs, splits or model settings.

Reproducing this published result requires matching its model configuration, data, split and scorer. Source checking or a successful smoke test does not establish score reproduction.

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.

20 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-10-23-5347695500ed
Property and statement	Original source and location	Review and provenance
Comparison: adaptation Supervised MLP on LLM embeddings, trained on BioGRID-ORCS Context-only references	Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? Original source ↗ Table 2, row "Summ. Emb." Version: Proceedings of the 24th Workshop on Biomedical Language Processing (BioNLP 2025), pages 354-364, 1 August 2025; ACL Anthology 2025.bionlp-1.30 Retrieved: 2026-10-22T13:57:26Z	not individually reviewed No individual claim review recorded author reported Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.comparison.adaptation` Source artifact SHA-256: `6c53ffec88444768fbb5c1a65031c096f07b3d9ac48de8163bc8f47d0d51205c` Hash scope: pdftotext -layout text layer, parsed by data/omics/use-case-refresh-phenotype-20261022/extract/song2025virtualcrispr.py with row labels asserted Inspected artifact
Comparison: aggregation Single trained model; one evaluation Context-only references	Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? Original source ↗ Table 2, row "Summ. Emb." Version: Proceedings of the 24th Workshop on Biomedical Language Processing (BioNLP 2025), pages 354-364, 1 August 2025; ACL Anthology 2025.bionlp-1.30 Retrieved: 2026-10-22T13:57:26Z	not individually reviewed No individual claim review recorded author reported Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.comparison.aggregation` Source artifact SHA-256: `6c53ffec88444768fbb5c1a65031c096f07b3d9ac48de8163bc8f47d0d51205c` Hash scope: pdftotext -layout text layer, parsed by data/omics/use-case-refresh-phenotype-20261022/extract/song2025virtualcrispr.py with row labels asserted Inspected artifact
Comparison: budget Not reported Context-only references	Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? Original source ↗ Table 2, row "Summ. Emb." Version: Proceedings of the 24th Workshop on Biomedical Language Processing (BioNLP 2025), pages 354-364, 1 August 2025; ACL Anthology 2025.bionlp-1.30 Retrieved: 2026-10-22T13:57:26Z	missing or unspecified No individual claim review recorded author reported Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.comparison.budget` Source artifact SHA-256: `6c53ffec88444768fbb5c1a65031c096f07b3d9ac48de8163bc8f47d0d51205c` Hash scope: pdftotext -layout text layer, parsed by data/omics/use-case-refresh-phenotype-20261022/extract/song2025virtualcrispr.py with row labels asserted Inspected artifact
Comparison: dataset version Curated from PMID 39567689 and PMID 39385035 Context-only references	Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? Original source ↗ Table 2, row "Summ. Emb." Version: Proceedings of the 24th Workshop on Biomedical Language Processing (BioNLP 2025), pages 354-364, 1 August 2025; ACL Anthology 2025.bionlp-1.30 Retrieved: 2026-10-22T13:57:26Z	not individually reviewed No individual claim review recorded author reported Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.comparison.dataset_version` Source artifact SHA-256: `6c53ffec88444768fbb5c1a65031c096f07b3d9ac48de8163bc8f47d0d51205c` Hash scope: pdftotext -layout text layer, parsed by data/omics/use-case-refresh-phenotype-20261022/extract/song2025virtualcrispr.py with row labels asserted Inspected artifact
Comparison: inputs Text of the perturbation method, gene symbol, cell line and phenotype, each embedded with text-embedding-3-large (raw text, or a GPT-4o summary for Summ. Emb.) and concatenated Context-only references	Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? Original source ↗ Table 2, row "Summ. Emb." Version: Proceedings of the 24th Workshop on Biomedical Language Processing (BioNLP 2025), pages 354-364, 1 August 2025; ACL Anthology 2025.bionlp-1.30 Retrieved: 2026-10-22T13:57:26Z	not individually reviewed No individual claim review recorded author reported Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.comparison.inputs` Source artifact SHA-256: `6c53ffec88444768fbb5c1a65031c096f07b3d9ac48de8163bc8f47d0d51205c` Hash scope: pdftotext -layout text layer, parsed by data/omics/use-case-refresh-phenotype-20261022/extract/song2025virtualcrispr.py with row labels asserted Inspected artifact
Comparison: metric implementation AUROC and AUPRC from predicted probabilities; other metrics after binarising with Youden's J statistic Context-only references	Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? Original source ↗ Table 2, row "Summ. Emb." Version: Proceedings of the 24th Workshop on Biomedical Language Processing (BioNLP 2025), pages 354-364, 1 August 2025; ACL Anthology 2025.bionlp-1.30 Retrieved: 2026-10-22T13:57:26Z	not individually reviewed No individual claim review recorded author reported Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.comparison.metric_implementation` Source artifact SHA-256: `6c53ffec88444768fbb5c1a65031c096f07b3d9ac48de8163bc8f47d0d51205c` Hash scope: pdftotext -layout text layer, parsed by data/omics/use-case-refresh-phenotype-20261022/extract/song2025virtualcrispr.py with row labels asserted Inspected artifact
Comparison: population 1,814 gene-phenotype queries, 907 positive and 907 negative Context-only references	Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? Original source ↗ Table 2, row "Summ. Emb." Version: Proceedings of the 24th Workshop on Biomedical Language Processing (BioNLP 2025), pages 354-364, 1 August 2025; ACL Anthology 2025.bionlp-1.30 Retrieved: 2026-10-22T13:57:26Z	not individually reviewed No individual claim review recorded author reported Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.comparison.population` Source artifact SHA-256: `6c53ffec88444768fbb5c1a65031c096f07b3d9ac48de8163bc8f47d0d51205c` Hash scope: pdftotext -layout text layer, parsed by data/omics/use-case-refresh-phenotype-20261022/extract/song2025virtualcrispr.py with row labels asserted Inspected artifact
Comparison: protocol id pheno-20261022-protocol-song2025-difficult-hit-classification Context-only references	Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? Original source ↗ Table 2, row "Summ. Emb." Version: Proceedings of the 24th Workshop on Biomedical Language Processing (BioNLP 2025), pages 354-364, 1 August 2025; ACL Anthology 2025.bionlp-1.30 Retrieved: 2026-10-22T13:57:26Z	not individually reviewed No individual claim review recorded author reported Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.comparison.protocol_id` Source artifact SHA-256: `6c53ffec88444768fbb5c1a65031c096f07b3d9ac48de8163bc8f47d0d51205c` Hash scope: pdftotext -layout text layer, parsed by data/omics/use-case-refresh-phenotype-20261022/extract/song2025virtualcrispr.py with row labels asserted Inspected artifact
Comparison: split Fixed evaluation set of 1,814 queries; no data from BioGRID-ORCS Context-only references	Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? Original source ↗ Table 2, row "Summ. Emb." Version: Proceedings of the 24th Workshop on Biomedical Language Processing (BioNLP 2025), pages 354-364, 1 August 2025; ACL Anthology 2025.bionlp-1.30 Retrieved: 2026-10-22T13:57:26Z	not individually reviewed No individual claim review recorded author reported Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.comparison.split` Source artifact SHA-256: `6c53ffec88444768fbb5c1a65031c096f07b3d9ac48de8163bc8f47d0d51205c` Hash scope: pdftotext -layout text layer, parsed by data/omics/use-case-refresh-phenotype-20261022/extract/song2025virtualcrispr.py with row labels asserted Inspected artifact
Limitations 3 values Authors' own method. F1, FPR, PPV, NPV, sensitivity and specificity use a Youden's J threshold; the paper does not say whether it was chosen on the validation screens or on the test queries. If on the test queries these values are optimistic. AUROC and AUPRC do not depend on the threshold. Trained on BioGRID-ORCS author hit calls; scoring it on other ORCS-derived benchmarks (for example AssayBench) would need a screen-level overlap check against its training screens. Context-only references	Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? Original source ↗ Table 2, row "Summ. Emb." Version: Proceedings of the 24th Workshop on Biomedical Language Processing (BioNLP 2025), pages 354-364, 1 August 2025; ACL Anthology 2025.bionlp-1.30 Retrieved: 2026-10-22T13:57:26Z	not individually reviewed No individual claim review recorded author reported Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.limitations` Source artifact SHA-256: `6c53ffec88444768fbb5c1a65031c096f07b3d9ac48de8163bc8f47d0d51205c` Hash scope: pdftotext -layout text layer, parsed by data/omics/use-case-refresh-phenotype-20261022/extract/song2025virtualcrispr.py with row labels asserted Inspected artifact

How to read the evidence

Sources and history

Release 2026-10-23-5347695500ed · Record review: source checked

1 source record and release history

Virtual CRISPR: Can LLMs Predict CRISPR Screen Results? · Original source · Proceedings of the 24th Workshop on Biomedical Language Processing (BioNLP 2025), pages 354-364, 1 August 2025; ACL Anthology 2025.bionlp-1.30

Download this release (gzip)

Technical metadata and extraction receipts

Stable ID: pheno-20261022-eval-song2025-embedding-classifier-summary-difficult

areas: cells-tissues
contexts: research
origin: author_reported
protocol: pheno-20261022-protocol-song2025-difficult-hit-classification
version: Primary source as retrieved 2026-10-22
comparison: protocol id: pheno-20261022-protocol-song2025-difficult-hit-classification; dataset version: Curated from PMID 39567689 and PMID 39385035; split: Fixed evaluation set of 1,814 queries; no data from BioGRID-ORCS; population: 1,814 gene-phenotype queries, 907 positive and 907 negative; inputs: Text of the perturbation method, gene symbol, cell line and phenotype, each embedded with text-embedding-3-large (raw text, or a GPT-4o summary for Summ. Emb.) and concatenated; metric implementation: AUROC and AUPRC from predicted probabilities; other metrics after binarising with Youden's J statistic; budget: Not reported; adaptation: Supervised MLP on LLM embeddings, trained on BioGRID-ORCS; aggregation: Single trained model; one evaluation
source locator: Table 2, row "Summ. Emb."
retrospective: true
limitations: Authors' own method.; F1, FPR, PPV, NPV, sensitivity and specificity use a Youden's J threshold; the paper does not say whether it was chosen on the validation screens or on the test queries. If on the test queries these values are optimistic. AUROC and AUPRC do not depend on the threshold.; Trained on BioGRID-ORCS author hit calls; scoring it on other ORCS-derived benchmarks (for example AssayBench) would need a screen-level overlap check against its training screens.

Related records

system: Virtual CRISPR embedding classifier, GPT-4o summary embeddings (Song et al. 2025)
assessment: Virtual CRISPR Difficult Benchmark: hit prediction for complex phenotypes
data: Virtual CRISPR Difficult Benchmark
evaluation: song2025-embedding-classifier-summary-difficult auprc
evaluation: song2025-embedding-classifier-summary-difficult auroc
evaluation: song2025-embedding-classifier-summary-difficult f1-score
evaluation: song2025-embedding-classifier-summary-difficult false-positive-rate
evaluation: song2025-embedding-classifier-summary-difficult negative-predictive-value
evaluation: song2025-embedding-classifier-summary-difficult precision
evaluation: song2025-embedding-classifier-summary-difficult recall
evaluation: song2025-embedding-classifier-summary-difficult specificity

Suggest a correction