rewirebio.iobenchmarks
Configuration

PerturbPFN step-3100 checkpoint, in-context forward pass

Configures PerturbPFN as the step-3100 checkpoint applied with one forward pass and no parameter updates, extracting a hard graph by prior-density calibration and greedy DAG projection and gating targets with a top-3 straight-through gate.

2 evaluations · 17 results

Overview

An explanatory profile has not yet been reviewed for this record.

Consult the linked sources for architecture or protocol details. Missing evidence is not evidence of a missing capability.

Model Capabilities & Hardware Requirements

Deployment facts recorded for PerturbPFN step-3100 checkpoint, in-context forward pass. Fields the record does not carry are marked not recorded.

ParametersNot recorded
Minimum VRAMNot recorded
Inference LatencyNot recorded
Context WindowNot recorded
AccessNot recorded
LicenseNot recorded

Evaluated Checkpoint: Step 3100, as reported in arXiv:2607.23447v1

Evaluations and results

2 evaluations · 17 results. Different protocols are not a single leaderboard.

Sorted by AUPRC (Intervention targets: node-wise target probabilities; PerturbPFN in-prior synthetic episodes) (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 configurationProtocol and datasetAUPRC (Intervention targets: node-wise target probabilities; PerturbPFN in-prior synthetic episodes)AUPRC (Latent graph: all directed non-self edges ranked by predicted edge probability, edge prevalence 0.0363; PerturbPFN in-prior synthetic episodes)AUROC (Intervention targets: node-wise target probabilities; PerturbPFN in-prior synthetic episodes)AUROC (Latent graph: all directed non-self edges ranked by predicted edge probability; PerturbPFN in-prior synthetic episodes)Cosine similarity (Predicted against true mean response shift relative to control, both centred across genes; PerturbPFN in-prior synthetic episodes)F1 (Intervention targets: node-wise probabilities thresholded at 0.5; PerturbPFN in-prior synthetic episodes)F1 (Intervention targets: straight-through top-k gate used in the decoder rollout (k = 3, Table 6); PerturbPFN in-prior synthetic episodes)F1 (Latent graph edges after keeping the top-scoring acyclic edges up to the true edge count (density-matched DAG projection); PerturbPFN in-prior synthetic episodes)Min max normalised score (Debiased Sinkhorn 2-Wasserstein distance between predicted and observed cell sets (epsilon 0.1), negated, then min-max normalised across the seven Table 2 rows within each of 12 Sci-Plex highest-dose protocols and averaged over protocols)Min max normalised score (Mean distance (L2 distance between predicted and observed mean expression over the 50 marker genes), negated, then min-max normalised across the seven Table 2 rows within each of 12 Sci-Plex highest-dose protocols and averaged over protocols)Min max normalised score (Pearson correlation between predicted and observed mean expression over the 50 marker genes, min-max normalised across the seven Table 2 rows within each of 12 Sci-Plex highest-dose protocols and averaged over protocols)Negative log likelihood (Mean negative log-likelihood of query responses under the predicted zero-inflated log-normal distribution; PerturbPFN in-prior synthetic episodes)Rank (Rank of the overall normalised score among the seven Table 2 rows on 12 Sci-Plex highest-dose protocols)Runtime (Mean wall-clock seconds per Sci-Plex highest-dose protocol on one NVIDIA RTX 4090 worker, plus PerturbPFN's one-time pretraining cost divided across the 12 protocols)Runtime (Mean wall-clock seconds per Sci-Plex highest-dose protocol on one NVIDIA RTX 4090 worker, without any one-time pretraining cost)Weighted composite score (Equal-weight mean of the three protocol-averaged min-max normalised scores (mean distance, Pearson r, Sinkhorn W2) across the seven Table 2 rows on 12 Sci-Plex highest-dose protocols)Magnitude ratio (Mean absolute predicted intervention strength over mean absolute true strength on true target nodes; 1 is the correct scale; PerturbPFN in-prior synthetic episodes)
PerturbPFN step-3100 checkpoint, in-context forward passEvaluationPerturbPFN in-prior latent recovery on held-out synthetic episodes (Table 1), PerturbPFN synthetic prior, 1,024 held-out episodes0.7860.2980.9330.8130.7440.7650.6290.326Not reportedNot reportedNot reported-0.138Not reportedNot reportedNot reportedNot reported0.939
PerturbPFN step-3100 checkpoint, in-context forward passEvaluationSci-Plex highest-dose holdout with min-max normalised MD, Pearson r and W2 (PerturbPFN Table 2), Sci-Plex HDAC-inhibitor subset, 3 cell lines by 4 drugs (GIM preprocessing)Not reportedNot reportedNot reportedNot reportedNot reportedNot reportedNot reportedNot reported0.9180.7190.881Not reported(#2)4331.110.839Not reported
All 17 result rows with coverage, uncertainty and sources
Filter evaluations

Applied filters: All linked evaluations

Exact evaluated configurations and original reported results
Tested configurationProtocol and datasetFindingEvidence and details
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: PerturbPFN in-prior latent recovery on held-out synthetic episodes (Table 1)
Dataset: PerturbPFN synthetic prior, 1,024 held-out episodes
0.298 AUPRC (Latent graph: all directed non-self edges ranked by predicted edge probability, edge prevalence 0.0363; PerturbPFN in-prior synthetic episodes)
unitless · higher

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on in-prior latent recovery (PerturbPFN Table 1)

perturbpfn-20261115-protocol-in-prior-latent-recovery

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 1, block 'Graph', row "AUPRC ↑"
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: PerturbPFN in-prior latent recovery on held-out synthetic episodes (Table 1)
Dataset: PerturbPFN synthetic prior, 1,024 held-out episodes
0.813 AUROC (Latent graph: all directed non-self edges ranked by predicted edge probability; PerturbPFN in-prior synthetic episodes)
unitless · higher

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on in-prior latent recovery (PerturbPFN Table 1)

perturbpfn-20261115-protocol-in-prior-latent-recovery

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 1, block 'Graph', row "AUROC ↑"
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: PerturbPFN in-prior latent recovery on held-out synthetic episodes (Table 1)
Dataset: PerturbPFN synthetic prior, 1,024 held-out episodes
0.326 F1 (Latent graph edges after keeping the top-scoring acyclic edges up to the true edge count (density-matched DAG projection); PerturbPFN in-prior synthetic episodes)
unitless · higher

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on in-prior latent recovery (PerturbPFN Table 1)

perturbpfn-20261115-protocol-in-prior-latent-recovery

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 1, block 'Graph', row "Proj. F1 ↑"
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: PerturbPFN in-prior latent recovery on held-out synthetic episodes (Table 1)
Dataset: PerturbPFN synthetic prior, 1,024 held-out episodes
-0.138 Negative log likelihood (Mean negative log-likelihood of query responses under the predicted zero-inflated log-normal distribution; PerturbPFN in-prior synthetic episodes)
unit-unreported · lower

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on in-prior latent recovery (PerturbPFN Table 1)

perturbpfn-20261115-protocol-in-prior-latent-recovery

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 1, block 'Strength / Response', row "Obs. NLL ↓"
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: PerturbPFN in-prior latent recovery on held-out synthetic episodes (Table 1)
Dataset: PerturbPFN synthetic prior, 1,024 held-out episodes
0.744 Cosine similarity (Predicted against true mean response shift relative to control, both centred across genes; PerturbPFN in-prior synthetic episodes)
unitless · higher

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on in-prior latent recovery (PerturbPFN Table 1)

perturbpfn-20261115-protocol-in-prior-latent-recovery

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 1, block 'Strength / Response', row "∆ cosine ↑"
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: PerturbPFN in-prior latent recovery on held-out synthetic episodes (Table 1)
Dataset: PerturbPFN synthetic prior, 1,024 held-out episodes
0.939 Magnitude ratio (Mean absolute predicted intervention strength over mean absolute true strength on true target nodes; 1 is the correct scale; PerturbPFN in-prior synthetic episodes)
unitless · unknown

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on in-prior latent recovery (PerturbPFN Table 1)

perturbpfn-20261115-protocol-in-prior-latent-recovery

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 1, block 'Strength / Response', row "|ψ| ratio →1"
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: PerturbPFN in-prior latent recovery on held-out synthetic episodes (Table 1)
Dataset: PerturbPFN synthetic prior, 1,024 held-out episodes
0.786 AUPRC (Intervention targets: node-wise target probabilities; PerturbPFN in-prior synthetic episodes)
unitless · higher

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on in-prior latent recovery (PerturbPFN Table 1)

perturbpfn-20261115-protocol-in-prior-latent-recovery

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 1, block 'Target', row "AUPRC ↑"
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: PerturbPFN in-prior latent recovery on held-out synthetic episodes (Table 1)
Dataset: PerturbPFN synthetic prior, 1,024 held-out episodes
0.933 AUROC (Intervention targets: node-wise target probabilities; PerturbPFN in-prior synthetic episodes)
unitless · higher

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on in-prior latent recovery (PerturbPFN Table 1)

perturbpfn-20261115-protocol-in-prior-latent-recovery

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 1, block 'Target', row "AUROC ↑"
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: PerturbPFN in-prior latent recovery on held-out synthetic episodes (Table 1)
Dataset: PerturbPFN synthetic prior, 1,024 held-out episodes
0.765 F1 (Intervention targets: node-wise probabilities thresholded at 0.5; PerturbPFN in-prior synthetic episodes)
unitless · higher

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on in-prior latent recovery (PerturbPFN Table 1)

perturbpfn-20261115-protocol-in-prior-latent-recovery

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 1, block 'Target', row "F1 ↑"
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: PerturbPFN in-prior latent recovery on held-out synthetic episodes (Table 1)
Dataset: PerturbPFN synthetic prior, 1,024 held-out episodes
0.629 F1 (Intervention targets: straight-through top-k gate used in the decoder rollout (k = 3, Table 6); PerturbPFN in-prior synthetic episodes)
unitless · higher

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on in-prior latent recovery (PerturbPFN Table 1)

perturbpfn-20261115-protocol-in-prior-latent-recovery

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Section 4.1 'Results', sentence 'The straight-through top-k gate used by the decoder obtains a lower F1 of 0.6287'
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: Sci-Plex highest-dose holdout with min-max normalised MD, Pearson r and W2 (PerturbPFN Table 2)
Dataset subset: Sci-Plex HDAC-inhibitor subset, 3 cell lines by 4 drugs (GIM preprocessing)
0.839 Weighted composite score (Equal-weight mean of the three protocol-averaged min-max normalised scores (mean distance, Pearson r, Sinkhorn W2) across the seven Table 2 rows on 12 Sci-Plex highest-dose protocols)
unitless · higher

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on the Sci-Plex highest-dose holdout (PerturbPFN Table 2)

perturbpfn-20261115-protocol-sciplex-highest-dose-normalised

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 2, unlabelled score line printed between the rows 'PerturbPFN (amortized)' and 'PerturbPFN (ICL only)', column 'Avg. ŝ / rank', score before the bracket
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: Sci-Plex highest-dose holdout with min-max normalised MD, Pearson r and W2 (PerturbPFN Table 2)
Dataset subset: Sci-Plex HDAC-inhibitor subset, 3 cell lines by 4 drugs (GIM preprocessing)
(#2) Rank (Rank of the overall normalised score among the seven Table 2 rows on 12 Sci-Plex highest-dose protocols)
unitless · lower

Uncertainty: Not reported

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on the Sci-Plex highest-dose holdout (PerturbPFN Table 2)

perturbpfn-20261115-protocol-sciplex-highest-dose-normalised

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 2, unlabelled score line printed between the rows 'PerturbPFN (amortized)' and 'PerturbPFN (ICL only)', column 'Avg. ŝ / rank', bracketed rank
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: Sci-Plex highest-dose holdout with min-max normalised MD, Pearson r and W2 (PerturbPFN Table 2)
Dataset subset: Sci-Plex HDAC-inhibitor subset, 3 cell lines by 4 drugs (GIM preprocessing)
433 Runtime (Mean wall-clock seconds per Sci-Plex highest-dose protocol on one NVIDIA RTX 4090 worker, plus PerturbPFN's one-time pretraining cost divided across the 12 protocols)
second · lower

Uncertainty: SD 0.34

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on the Sci-Plex highest-dose holdout (PerturbPFN Table 2)

perturbpfn-20261115-protocol-sciplex-highest-dose-normalised

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 2, row 'PerturbPFN (amortized)', column 'Time (s) / protocol'
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: Sci-Plex highest-dose holdout with min-max normalised MD, Pearson r and W2 (PerturbPFN Table 2)
Dataset subset: Sci-Plex HDAC-inhibitor subset, 3 cell lines by 4 drugs (GIM preprocessing)
1.11 Runtime (Mean wall-clock seconds per Sci-Plex highest-dose protocol on one NVIDIA RTX 4090 worker, without any one-time pretraining cost)
second · lower

Uncertainty: SD 0.34

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on the Sci-Plex highest-dose holdout (PerturbPFN Table 2)

perturbpfn-20261115-protocol-sciplex-highest-dose-normalised

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 2, row 'PerturbPFN (ICL only)', column 'Time (s) / protocol'
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: Sci-Plex highest-dose holdout with min-max normalised MD, Pearson r and W2 (PerturbPFN Table 2)
Dataset subset: Sci-Plex HDAC-inhibitor subset, 3 cell lines by 4 drugs (GIM preprocessing)
0.719 Min max normalised score (Mean distance (L2 distance between predicted and observed mean expression over the 50 marker genes), negated, then min-max normalised across the seven Table 2 rows within each of 12 Sci-Plex highest-dose protocols and averaged over protocols)
unitless · higher

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on the Sci-Plex highest-dose holdout (PerturbPFN Table 2)

perturbpfn-20261115-protocol-sciplex-highest-dose-normalised

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 2, unlabelled score line printed between the rows 'PerturbPFN (amortized)' and 'PerturbPFN (ICL only)', column 'ŝMD ↑'
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: Sci-Plex highest-dose holdout with min-max normalised MD, Pearson r and W2 (PerturbPFN Table 2)
Dataset subset: Sci-Plex HDAC-inhibitor subset, 3 cell lines by 4 drugs (GIM preprocessing)
0.881 Min max normalised score (Pearson correlation between predicted and observed mean expression over the 50 marker genes, min-max normalised across the seven Table 2 rows within each of 12 Sci-Plex highest-dose protocols and averaged over protocols)
unitless · higher

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on the Sci-Plex highest-dose holdout (PerturbPFN Table 2)

perturbpfn-20261115-protocol-sciplex-highest-dose-normalised

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 2, unlabelled score line printed between the rows 'PerturbPFN (amortized)' and 'PerturbPFN (ICL only)', column 'ŝr ↑'
Configuration: PerturbPFN step-3100 checkpoint, in-context forward passProtocol: Sci-Plex highest-dose holdout with min-max normalised MD, Pearson r and W2 (PerturbPFN Table 2)
Dataset subset: Sci-Plex HDAC-inhibitor subset, 3 cell lines by 4 drugs (GIM preprocessing)
0.918 Min max normalised score (Debiased Sinkhorn 2-Wasserstein distance between predicted and observed cell sets (epsilon 0.1), negated, then min-max normalised across the seven Table 2 rows within each of 12 Sci-Plex highest-dose protocols and averaged over protocols)
unitless · higher

Uncertainty: Not reported by the source: No spread is printed for this value.

Coverage: Not reported scored / Not reported eligible

Author-reported evaluation · Source checked
Methods, coverage and source

PerturbPFN on the Sci-Plex highest-dose holdout (PerturbPFN Table 2)

perturbpfn-20261115-protocol-sciplex-highest-dose-normalised

Aggregation: Not reported

PerturbPFN: Probing the Limits of Synthetic Priors in Drug Perturbation Modelling (arXiv:2607.23447v1) · Table 2, unlabelled score line printed between the rows 'PerturbPFN (amortized)' and 'PerturbPFN (ICL only)', column 'ŝW2 ↑'

Source checking is not independent reproduction. Release 2026-11-21-c8bea25e5abb.

Use this configuration

How it works, versions and access

Additional methodological details, versions and access terms have not yet been extracted into a structured profile for this configuration. Consult the linked sources and evaluation records below.

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

Claims, original sources and review scope · Release 2026-11-21-c8bea25e5abb
Property and statementOriginal source and locationReview and provenance

No evidence rows match these filters. Choose another scope or clear the search.

Sources and history

Release 2026-11-21-c8bea25e5abb · Record review: source checked

1 source record and release historyDownload this release (gzip)
Technical metadata and record identifiers

Technical metadata for this record is tracked under stable ID perturbpfn-20261115-config-perturbpfn-step3100.

Source locator: Sections 2.4, 2.5 and 4; Appendix B; Tables 6 and 7

Related records

How to cite

PerturbPFN step-3100 checkpoint, in-context forward pass. Rewire biological model benchmark database, release 2026-11-21-c8bea25e5abb. https://benchmarks.rewirebio.io/database/configuration/perturbpfn-20261115-config-perturbpfn-step3100/

Release 2026-11-21-c8bea25e5abb. Rankings and values change between releases, so keep the release id in the citation. Cite the original publication as well for any value you reuse: each result links its source.

The compilation and annotations made by Rewire in this release are offered under CC BY 4.0; values and text taken from cited sources stay under the terms of those sources and are not covered by this licence. Data licence

Suggest a correction