What the evidence shows
AgroNT scored above the task-specific CNN in all six assay-by-species comparisons, but by 0.02 to 0.05 R-squared, with no uncertainty reported.
On STARR-seq core-promoter strength, AgroNT reached R-squared 0.62 to 0.75 and the CNN from Jores et al. 0.57 to 0.73, across maize protoplasts and tobacco leaves for Arabidopsis thaliana, Sorghum bicolor and Zea mays sequences. In the earlier Jores study, which pools the three species, the CNN reached a squared Pearson correlation of 0.67 in maize protoplasts and 0.71 in tobacco leaves, against 0.45 and 0.51 for a GC and motif linear model. In an independent benchmark, five DNA foundation models separate Arabidopsis TATA and non-TATA promoter sequences from similar non-promoter sequences at AUC 0.937 to 0.961, but that is sequence classification, not reporter strength, and it uses a different dataset.
The AgroNT and Jores values are reported by the authors of those studies, the six AgroNT comparisons all come from one figure of the AgroNT paper, and no value has a confidence interval or run variability. Nothing here covers stable transformed plants, tissue specificity or newly designed promoters.
- You bring
- A 170 bp core-promoter sequence spanning −165 to +5 relative to an annotated transcription start site
- The sequence species and intended assay host: maize protoplasts or tobacco leaves
- You want
- Assay-specific evidence for AgroNT and the CNN from Jores et al., with exact evaluated configurations and source rows; no general plant-performance ranking.
- Evidence in this release
- 10 comparisons shown · 10 tools · 18 configurations
- Kept up to date
- Evidence last checked 8 October 2026 · re-checked monthly, next due 8 November 2026
For "Compare methods for plant promoter experiments", on Feng Promoter Arabidopsis NonTATA sequence classification, AUROC (task: promoter arabidopsis non-tata) (AUROC, a proxy protocol), HyenaDNA scored 0.9547, reported by Benchmarking DNA foundation models for genomic and genetic tasks.
Best reported valueHyenaDNABenchmarking DNA foundation models for genomic and genetic…
AUROC0.9547Feng Promoter Arabidopsis NonTATA sequence classification, AUROC (task: promoter arabidopsis non-tata)
Baseline on this protocolNone scoredNo margin is computed
Ranked tools55 from independent studies
5 tools have a value on this protocol; 5 of them come from studies whose authors did not develop the tool. No baseline method is scored on this protocol. Reviewers judged this protocol a proxy for the use case.
Ranking
Ranked on Feng Promoter Arabidopsis NonTATA sequence classification, AUROC (task: promoter arabidopsis non-tata) (AUROC, higher is better). Proxy protocol Reviewers judged this protocol a proxy for the use case, not a direct test of it. Only results on this protocol are ranked; other tools are listed below the table.
5 tools, from 1 source. 5 of 5 results (100%) come from an independent study.
10 tools are evaluated elsewhere in this use case but not on this protocol, so not ranked
Baseline
GC and motif linear regression. Linear regression on GC content and core-promoter and transcription-factor motif scores is the simple model the plant STARR-seq study fitted before turning to a CNN. It is a simple statistical model fitted to the training promoters; no other established predictor of plant promoter strength is measured in the linked evidence.
Machine-learning methods are compared with the baseline on the same data in 2 tables: Jores 2021 promoter strength: maize-protoplasts, Jores 2021 promoter strength: tobacco-leaves. Baseline rows are marked in each table.
Comparisons
Unified model comparison for this use case. Compare tools against protocol baselines in a single view.
Proxy evidenceIndependent study: Reported in a study whose authors did not develop the evaluated tool.
Promoter Arabidopsis NonTATA sequence-classification AUC for five DNA foundation-model embeddings
Sorted by AUROC (higher is better). The best value in each column is highlighted.
| Tool | AUROC (task: promoter arabidopsis non-tata) |
|---|
| 0.9547 (best in column) |
|---|
| 0.9490 |
|---|
| 0.9457 |
|---|
| 0.9437 |
|---|
| 0.9395 |
|---|
Caveats and method
- Sequence classification accuracy is not a reporter expression measurement.
- Exact per-task split sizes/counts are unextracted from the inspected main text (Supplementary Data 6 is identified but not inspected).
- No independent reproduction or qualified human scientific review.
Protocol: Feng Promoter Arabidopsis NonTATA sequence classification
Primary benchmark evaluation shown (Feng Promoter Arabidopsis NonTATA sequence classification). 9 additional study protocols (Jores 2021 promoter strength: maize-protoplasts, Jores 2021 promoter strength: tobacco-leaves, Feng Promoter Arabidopsis TATA sequence classification and others) are cited in individual tool drawer rows and under Question, sources and review.
Not covered yet
- All values are author-reported and source checked; no independent experimental reproduction or human scientific review is recorded.
- The source tables provide no confidence intervals or run variability. Exact fitted checkpoint hashes, task-specific seeds, scoring denominators and executable split manifests remain unreported or unextracted.
- The six comparisons retain their own assay system and sequence species. A maize-protoplast model can be evaluated on sequences from any of the three listed species; the host is not the sequence origin.
- Held-out labelled examples do not establish exclusion from AgroNT pretraining or generalisation to an unseen species. Its reference-genome pretraining and supervised assay split answer different leakage questions.
- Held-out R² describes assay-strength prediction. It is not top-k candidate hit rate, uncertainty for a new promoter, or proof that a designed promoter will work in a stable plant.
- These four task-specific configurations do not establish performance for other AgroNT checkpoints or newer methods. Missing checkpoint and fitting detail limits exact recreation of the reported comparison.
- The separately flagged Figure 4 chromatin-accessibility source anomaly is outside this page and supplies no evidence for promoter selection.
- 30 September 2026 audit: Reporter-host and sequence-species conditions cannot be pooled. Jores pooled-species Fig8 and AgroNT species-specific Fig3e protocols remain separate.
- 30 September 2026 audit: Prospective experimental promoter evolution is described by Jores; no benchmarked prospective model-to-model design hit-rate or stable-plant endpoint has been extracted.
- 30 September 2026 audit: Uncertainty, exact test counts and split/checkpoint hashes remain unextracted for the newly added historical comparison.
Contribute evidence or propose a correction
How to cite
Compare methods for plant promoter experiments. Rewire biological model benchmark database, release 2026-11-21-c8bea25e5abb. https://benchmarks.rewirebio.io/use-cases/plant-promoter-reporters/
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
Question, sources and review
Who this is for, setting and exclusions
Choose the matching sequence species and reporter host, then inspect the two evaluated methods and remaining validation needs before prioritising new promoter candidates. Keep each assay and species comparison separate.
- Who this is for
- Plant synthetic-biology researchers prioritising promoter candidates for reporter experiments
- Computational researchers selecting an assay-matched promoter prediction method
- Biological setting
- The evidence concerns held-out core-promoter sequences from Arabidopsis thaliana, Sorghum bicolor and Zea mays. Their activity was measured by transient STARR-seq reporter assays in maize protoplasts or tobacco leaves, using the original studies’ train/test datasets for the model comparisons.
- Research setting
- This page concerns plant reporter research. The evidence establishes no clinical application.
Outside this use case
- Endogenous gene expression, stable transformed plants, tissue-specific activity outside the two reporter hosts, plant fitness or crop yield
- Terminator strength, promoter–terminator combinations, chromatin accessibility and other AgroNT tasks
- Prospective validation of newly designed promoters, or pooling the six assay-by-species conditions into one score
Related articles
Read the background, methods and practical considerations behind this question.
Explains how species, sequence context and assay endpoints constrain the use of genomic models for regulatory prediction.
Sources, review and downloads
Automated source review · 2026-09-30 · Codex source curation, independent worker cross-review and root integration review
Source-backed literature curation with independent automated transcription and scope checks. No new model execution, independent experimental replication, qualified human scientific review or clinical validation.
Release 2026-11-21-c8bea25e5abb · question use-case-plant-promoter-reporters
Use-case input digest bfbb7ab9303becad1e3cb395568c26f41e72ddb67cd1a89a06d9eaa0f76bd95f
Download questions and review metadata (JSON) (gzip) · Verify release checksums (gzip)