Pipeline

scGPT + residual geometry

This pipeline tests whether scGPT residual-stream geometry adds signal for gene-regulatory links beyond expression features.

Sources

Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1); Background (paragraph 7)

1 evaluation · 1 metric row

How it worksEvaluated procedure (conceptual)

Conceptual input–method–output guide. Check the procedure text and linked evaluation for fitted components, additional inputs and exact settings.

Sources

Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · Methods/Geometric feature extraction (paragraph 2); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1)

At a glance

Model type

Single-cell foundation model; this record is the paper-specific evaluated configuration.

Sources

bowang-lab/scGPT README.md · README.md model description

Biological inputs

Tabula Sapiens single-cell expression and curated TRRUST TF–target edges

Sources

Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · Methods/Data sources and preprocessing (paragraph 1); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1)

Outputs

Regulatory-edge prediction scores and incremental performance over expression controls

Sources

Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · Methods/Ensembling with expression-based GRN inference (paragraph 1); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1)

limited source coverage · Automated source review, 2026-09-16. All specifications and missing details

Evaluations and results

Release 2026-09-17-d277315f7d76 · 1 evaluation · 1 metric row. Different protocols are not a single leaderboard.

Evaluation setupMetricEvidence origin

Results grouped by the exact reported evaluation
Metric and finding	Coverage and uncertainty	Evidence
scGPT + residual geometry: gene-regulatory signal prediction Pipeline: scGPT + residual geometryTask: gene-regulatory signal predictionDataset: immune tissue Asymmetric extraction, PCA-64 centered cosine geometry added to scGPT baseline Author-reported evaluation · Evaluation metadata: needs review
0.677 AUROC Unit: fraction · Direction: unknown	Uncertainty: not reported in legacy extract Scored: Not reported · Eligible: Not reported	source checkedResidual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · Table 4, Immune row, scGPT > +geom AUROC column Source checking is not independent reproduction.

How it works

How the evaluated method works

Gene vectors extracted from frozen representations yield cosine, centred-cosine, PCA-projected or multilayer geometry features. Supervised comparison with expression controls tests incremental regulatory-link prediction.

Sources

Underlying method and version boundaries

The official scGPT implementation supplies pretrained checkpoints and separate workflows for embedding extraction, cell annotation, integration and perturbation modelling. The checkpoint and adaptation procedure must be identified separately for each result.

Sources

bowang-lab/scGPT README.md · README.md; introduction, model description, pretrained-model and usage sections at pinned revision

What was evaluated

The linked evaluation record identifies scGPT + residual geometry: gene-regulatory signal prediction. Its dataset, split, adaptation and evidence origin remain attached to the reported results.

Sources

Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · The named evaluation’s methods and comparison table; exact preserved evaluation IDs: evaluation-lit-b4-013

Strengths and limitations

Strengths and considerations

Includes label-permutation and geometry-shuffle controls to test whether added geometric features carry incremental signal.
Sources
Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · Results/Geometric signal is tissue-dependent but recoverable through methodological refinement (paragraph 2); Methods/Null controls (paragraph 2)

Limitations and conditions

Predictive association with curated links is not proof of causality, and tissue-specific effects must not be universalised.
Sources
Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · Methods/Edge classification framework (paragraph 1); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1)

Profile review details

Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction.

Stable record: reported-model-d6a7fa854437e8

Specifications

Inputs, training, access and other details

Explanatory profile: limited source coverage · Automated source review, 2026-09-16. Review applies to the cited claims; unresolved fields are listed below. Numerical results retain their own review status.

Inputs, outputs and configuration
Property	Description and evidence
Model type	Single-cell foundation model; this record is the paper-specific evaluated configuration. Sources bowang-lab/scGPT README.md · README.md model description
Architecture / procedure	Gene vectors extracted from frozen representations yield cosine, centred-cosine, PCA-projected or multilayer geometry features. Supervised comparison with expression controls tests incremental regulatory-link prediction. Sources Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · Methods/Geometric feature extraction (paragraph 2); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1)
Biological inputs	Tabula Sapiens single-cell expression and curated TRRUST TF–target edges Sources Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · Methods/Data sources and preprocessing (paragraph 1); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1)
Outputs	Regulatory-edge prediction scores and incremental performance over expression controls Sources Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · Methods/Ensembling with expression-based GRN inference (paragraph 1); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1)
Parameters	An aggregate parameter total for this exact evaluated configuration is not established by the inspected sources. · Not reported in inspected sources Sources (2) Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues; bowang-lab/scGPT README.md · Results/Geometric signal is tissue-dependent but recoverable through methodological refinement; Discussion/Complementarity between model architectures; Methods/Data sources and preprocessing; Methods/Foundation models; Methods/Geometric feature extraction; Methods/Edge classification framework; Methods/Evaluation protocol; Methods/Null controls; inspected for aggregate parameter count (component sizes are not added without an exact configuration); README.md at pinned repository revision
Known versions / configuration	scGPT + residual geometry is the comparison-table label; that label does not specify an immutable weight revision. · Not reported in inspected sources Sources Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · Model identification in the comparison table and corresponding Methods; immutable checkpoint revision is not supplied by the table label.
Training data / fitting	Four tissue contexts with repeated cross-validation, leave-TF-out/target-out tests and matched-negative controls. Sources Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · Conclusions (paragraph 1); Methods/Edge classification framework (paragraph 1)
Context limits	A maximum input/context length for this exact evaluated configuration is not established by the inspected sources. · Not reported in inspected sources Sources (2) Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues; bowang-lab/scGPT README.md · Results/Geometric signal is tissue-dependent but recoverable through methodological refinement; Discussion/Complementarity between model architectures; Methods/Data sources and preprocessing; Methods/Foundation models; Methods/Geometric feature extraction; Methods/Edge classification framework; Methods/Evaluation protocol; Methods/Null controls; inspected for explicit maximum input length (dataset lengths and family-wide limits are not substituted); README.md at pinned repository revision
Access	Official upstream implementation and usage documentation: https://github.com/bowang-lab/scGPT/blob/cebd6fae655b9c585a4807daa3ac31bb764f06b4/README.md. This pinned documentation revision is not automatically the evaluated weight revision. Sources bowang-lab/scGPT README.md · README.md; installation, model download and usage instructions
Code licence	MIT (upstream repository code at the cited revision; this does not establish every dependency or historical checkpoint licence). Sources bowang-lab/scGPT LICENSE · LICENSE; complete licence text
Weights licence	The inspected model-access documentation does not explicitly identify terms for this exact evaluated checkpoint or fitted head; repository code terms are shown separately. · Not reported in inspected sources Sources bowang-lab/scGPT README.md · README.md; checkpoint/access documentation and licence scope

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.

22 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-09-17-d277315f7d76
Property and statement	Original source and location	Review and provenance
Diagram caption Conceptual input–method–output guide. Check the procedure text and linked evaluation for fitted components, additional inputs and exact settings. Individual claims	Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues Original source ↗ Methods/Geometric feature extraction (paragraph 2); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1) Version: version of record Retrieved: 2026-09-16T10:41:06Z	source checked automated source review · 2026-09-16 Audit details Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: `attributes.profile.diagram.caption` Source artifact SHA-256: `77546faec51cfb5c78b73c5943f940c4e0097d130b1ddde4ab8287b507b36df6` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Diagram steps ["Tabula Sapiens single-cell expression and curated TRRUST TF–target edges","scGPT + residual geometry","Regulatory-edge prediction scores and incremental performance over expression controls"] Individual claims	Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues Original source ↗ Methods/Geometric feature extraction (paragraph 2); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1) Version: version of record Retrieved: 2026-09-16T10:41:06Z	source checked automated source review · 2026-09-16 Audit details Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: `attributes.profile.diagram.steps` Source artifact SHA-256: `77546faec51cfb5c78b73c5943f940c4e0097d130b1ddde4ab8287b507b36df6` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Diagram title Evaluated procedure (conceptual) Individual claims	Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues Original source ↗ Methods/Geometric feature extraction (paragraph 2); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1) Version: version of record Retrieved: 2026-09-16T10:41:06Z	source checked automated source review · 2026-09-16 Audit details Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: `attributes.profile.diagram.title` Source artifact SHA-256: `77546faec51cfb5c78b73c5943f940c4e0097d130b1ddde4ab8287b507b36df6` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Model type Single-cell foundation model; this record is the paper-specific evaluated configuration. Individual claims	bowang-lab/scGPT README.md Original source ↗ README.md model description Version: cebd6fae655b9c585a4807daa3ac31bb764f06b4 Retrieved: 2026-09-16T20:00:00.816587+00:00	source checked automated source review · 2026-09-16 Audit details Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: `attributes.profile.facts.0.value` Source artifact SHA-256: `b0503e8ca789f19f1fc2350c5aaf57b1b323bbae43b354655231b5f4a1586c83` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Architecture / procedure Gene vectors extracted from frozen representations yield cosine, centred-cosine, PCA-projected or multilayer geometry features. Supervised comparison with expression controls tests incremental regulatory-link prediction. Individual claims	Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues Original source ↗ Methods/Geometric feature extraction (paragraph 2); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1) Version: version of record Retrieved: 2026-09-16T10:41:06Z	source checked automated source review · 2026-09-16 Audit details Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: `attributes.profile.facts.1.value` Source artifact SHA-256: `77546faec51cfb5c78b73c5943f940c4e0097d130b1ddde4ab8287b507b36df6` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Weights licence The inspected model-access documentation does not explicitly identify terms for this exact evaluated checkpoint or fitted head; repository code terms are shown separately. Individual claims	bowang-lab/scGPT README.md Original source ↗ README.md; checkpoint/access documentation and licence scope Version: cebd6fae655b9c585a4807daa3ac31bb764f06b4 Retrieved: 2026-09-16T20:00:00.816587+00:00	unreported automated source review · 2026-09-16 Audit details Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: `attributes.profile.facts.10.value` Source artifact SHA-256: `b0503e8ca789f19f1fc2350c5aaf57b1b323bbae43b354655231b5f4a1586c83` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Biological inputs Tabula Sapiens single-cell expression and curated TRRUST TF–target edges Individual claims	Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues Original source ↗ Methods/Data sources and preprocessing (paragraph 1); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1) Version: version of record Retrieved: 2026-09-16T10:41:06Z	source checked automated source review · 2026-09-16 Audit details Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: `attributes.profile.facts.2.value` Source artifact SHA-256: `77546faec51cfb5c78b73c5943f940c4e0097d130b1ddde4ab8287b507b36df6` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Outputs Regulatory-edge prediction scores and incremental performance over expression controls Individual claims	Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues Original source ↗ Methods/Ensembling with expression-based GRN inference (paragraph 1); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1) Version: version of record Retrieved: 2026-09-16T10:41:06Z	source checked automated source review · 2026-09-16 Audit details Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: `attributes.profile.facts.3.value` Source artifact SHA-256: `77546faec51cfb5c78b73c5943f940c4e0097d130b1ddde4ab8287b507b36df6` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Parameters An aggregate parameter total for this exact evaluated configuration is not established by the inspected sources. Individual claims	bowang-lab/scGPT README.md Original source ↗ Results/Geometric signal is tissue-dependent but recoverable through methodological refinement; Discussion/Complementarity between model architectures; Methods/Data sources and preprocessing; Methods/Foundation models; Methods/Geometric feature extraction; Methods/Edge classification framework; Methods/Evaluation protocol; Methods/Null controls; inspected for aggregate parameter count (component sizes are not added without an exact configuration); README.md at pinned repository revision Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: cebd6fae655b9c585a4807daa3ac31bb764f06b4 Retrieved: 2026-09-16T20:00:00.816587+00:00	unreported automated source review · 2026-09-16 Audit details Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: `attributes.profile.facts.4.value` Source artifact SHA-256: `b0503e8ca789f19f1fc2350c5aaf57b1b323bbae43b354655231b5f4a1586c83` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Parameters An aggregate parameter total for this exact evaluated configuration is not established by the inspected sources. Individual claims	Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues Original source ↗ Results/Geometric signal is tissue-dependent but recoverable through methodological refinement; Discussion/Complementarity between model architectures; Methods/Data sources and preprocessing; Methods/Foundation models; Methods/Geometric feature extraction; Methods/Edge classification framework; Methods/Evaluation protocol; Methods/Null controls; inspected for aggregate parameter count (component sizes are not added without an exact configuration); README.md at pinned repository revision Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: version of record Retrieved: 2026-09-16T10:41:06Z	unreported automated source review · 2026-09-16 Audit details Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: `attributes.profile.facts.4.value` Source artifact SHA-256: `77546faec51cfb5c78b73c5943f940c4e0097d130b1ddde4ab8287b507b36df6` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact

All evidence (CSV)All evidence (JSONL)How to read the evidence

Sources and history

Release 2026-09-17-d277315f7d76 · Record review: needs review

3 source records and release history

Residual-stream geometry of single-cell foundation models carries incremental gene-regulatory signal across tissues · Original source · version of record
bowang-lab/scGPT README.md · Original source · cebd6fae655b9c585a4807daa3ac31bb764f06b4
bowang-lab/scGPT LICENSE · Original source · cebd6fae655b9c585a4807daa3ac31bb764f06b4

Download this release

Technical metadata and extraction receipts

Stable ID: reported-model-d6a7fa854437e8

areas: cells-tissues
entity level: method
version: Not reported
reported name: scGPT + residual geometry
historical missing metadata: version: not_reported_in_legacy_extract; checkpoint revision: not_reported_in_legacy_extract; training data: not_reported_in_legacy_extract; licence: not_reported_in_legacy_extract
metadata review scope: historical_missing_metadata preserves the original discovery state. Current descriptive evidence and missingness are recorded in profile.facts; numerical-result review is separate.
legacy kinds: model
entity classification: review date: 2026-09-17; rationale: This record identifies a composed analysis workflow with separately identifiable upstream models, representations or tools and a downstream prediction/scoring procedure. Results belong to that complete composition rather than to an upstream model alone.; source ids: single-cell-residual-geometry-2026; evidence-reported-base-scgpt-readme-md; source locator: Methods/Geometric feature extraction (paragraph 2); Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1) | README.md model description | Results/Residual-stream geometry carries regulatory signal beyond expression confounds (paragraph 1); Background (paragraph 7); ambiguities: This is the paper-specific pipeline identity; unspecified component checkpoints or implementation versions are not inferred from its name.

Related records

model: scGPT + residual geometry: gene-regulatory signal prediction

Suggest a correction