ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl
length 100, Quality ppl. Scored with ppl on Generated sequences at fixed lengths. Unconditional sequence generation at a fixed length. Values are the mean and standard deviation over repeated runs.
Overview
length 100, Quality ppl. Scored with ppl on Generated sequences at fixed lengths. Unconditional sequence generation at a fixed length. Values are the mean and standard deviation over repeated runs.
Consult the linked sources for architecture or protocol details. Missing evidence is not evidence of a missing capability.
Evaluation design
Benchmarks bring together tasks and protocols. A task describes the biological question; a protocol defines a particular test.
Benchmarks
These source-backed links do not make different protocols or scores interchangeable.
Recorded evaluations
Each evaluation records what was tested and under which conditions.
- DPLM (650M) on ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl
- ESM3 (1.4B) on ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl
- EvoDiff on ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl
- Progen 2 (700M) on ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl
Run instructions
No runnable recipe has been reviewed for this task. Dataset access, model requirements, licences and compute requirements must be checked against its sources before execution.
A task describes a biological question. Choose a linked protocol to obtain concrete split and scoring instructions.
Published comparisons
Explore the results reported under one evaluation protocol. Each figure keeps its source, dataset and metric together; it is not a ranking across studies. The pooled view gathers every source table that reports the same metric and names what it does not hold constant.
ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl
ppl (score) · Lower values are better for this metric.
Every method ProteinBench reports on length 100, Quality ppl, scored with ppl on Generated sequences at fixed lengths.
Evaluation protocol · Generated sequences at fixed lengths (ProteinBench split)
- Progen 2 (700M) · Configuration · Author-reported evaluation8.28 ± 3.87
- EvoDiff · Configuration · Author-reported evaluation16.89 ± 1.04
- DPLM (650M) · Configuration · Author-reported evaluation6.21 ± 3.10
- ESM3 (1.4B) · Configuration · Author-reported evaluation14.79 ± 2.90
Source order is preserved. Plotted marks show point estimates; uncertainty, where reported, is retained in the printed values and table. Differences do not establish statistical significance.
proteinbench primary benchmark evidence · Table 4, column(length 100, Quality ppl ↓)Values, uncertainty and evidence
| Tested entity | Printed value | Uncertainty | Evidence |
|---|---|---|---|
| Progen 2 (700M) · Configuration | 8.28 ± 3.87 score | type: standard_deviation; value: 3.87 | Author-reported evaluation · source checkedproteinbench primary benchmark evidence · Table 4, row(Progen 2 (700M)), column(length 100, Quality ppl ↓) |
| EvoDiff · Configuration | 16.89 ± 1.04 score | type: standard_deviation; value: 1.04 | Author-reported evaluation · source checkedproteinbench primary benchmark evidence · Table 4, row(EvoDiff), column(length 100, Quality ppl ↓) |
| DPLM (650M) · Configuration | 6.21 ± 3.10 score | type: standard_deviation; value: 3.10 | Author-reported evaluation · source checkedproteinbench primary benchmark evidence · Table 4, row(DPLM (650M)), column(length 100, Quality ppl ↓) |
| ESM3 (1.4B) · Configuration | 14.79 ± 2.90 score | type: standard_deviation; value: 2.90 | Author-reported evaluation · source checkedproteinbench primary benchmark evidence · Table 4, row(ESM3 (1.4B)), column(length 100, Quality ppl ↓) |
Scope and limitations
- Author-reported numbers, source checked but not independently reproduced.
- Many of these metrics are better when lower, including scRMSD, perplexity and binding energy. The direction comes from the arrow in the paper's own header.
- Rows such as Native PDBs and RAbD are the natural reference the authors include for scale, not a competing method.
Source transcription and grouping reviewed by automated source review on 2026-09-18. These experiments were not independently reproduced by rewire.
Tested entities and results
Release 2026-09-17-134cd1815de8 · 4 evaluations · 4 metric rows. Different protocols are not a single leaderboard. Where several source tables report the same metric, the published comparisons above offer a pooled view that names what it does not hold constant.
| Metric and finding | Coverage and uncertainty | Evidence |
|---|---|---|
| DPLM (650M) on ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl Configuration: DPLM (650M)Task: ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality pplDataset subset: Generated sequences at fixed lengths (ProteinBench split) Unconditional sequence generation at a fixed length. Values are the mean and standard deviation over repeated runs. Author-reported evaluation · Evaluation metadata: source checked | ||
| 6.21 ± 3.10 ppl Unit: score · Direction: lower | Uncertainty: type: standard deviation; value: 3.10 Scored: Not reported · Eligible: Not reported | source checkedproteinbench primary benchmark evidence · Table 4, row(DPLM (650M)), column(length 100, Quality ppl ↓) Source checking is not independent reproduction. |
| ESM3 (1.4B) on ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl Configuration: ESM3 (1.4B)Task: ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality pplDataset subset: Generated sequences at fixed lengths (ProteinBench split) Unconditional sequence generation at a fixed length. Values are the mean and standard deviation over repeated runs. Author-reported evaluation · Evaluation metadata: source checked | ||
| 14.79 ± 2.90 ppl Unit: score · Direction: lower | Uncertainty: type: standard deviation; value: 2.90 Scored: Not reported · Eligible: Not reported | source checkedproteinbench primary benchmark evidence · Table 4, row(ESM3 (1.4B)), column(length 100, Quality ppl ↓) Source checking is not independent reproduction. |
| EvoDiff on ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl Configuration: EvoDiffTask: ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality pplDataset subset: Generated sequences at fixed lengths (ProteinBench split) Unconditional sequence generation at a fixed length. Values are the mean and standard deviation over repeated runs. Author-reported evaluation · Evaluation metadata: source checked | ||
| 16.89 ± 1.04 ppl Unit: score · Direction: lower | Uncertainty: type: standard deviation; value: 1.04 Scored: Not reported · Eligible: Not reported | source checkedproteinbench primary benchmark evidence · Table 4, row(EvoDiff), column(length 100, Quality ppl ↓) Source checking is not independent reproduction. |
| Progen 2 (700M) on ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl Configuration: Progen 2 (700M)Task: ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality pplDataset subset: Generated sequences at fixed lengths (ProteinBench split) Unconditional sequence generation at a fixed length. Values are the mean and standard deviation over repeated runs. Author-reported evaluation · Evaluation metadata: source checked | ||
| 8.28 ± 3.87 ppl Unit: score · Direction: lower | Uncertainty: type: standard deviation; value: 3.87 Scored: Not reported · Eligible: Not reported | source checkedproteinbench primary benchmark evidence · Table 4, row(Progen 2 (700M)), column(length 100, Quality ppl ↓) Source checking is not 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.
1 evidence row matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|---|---|
| Relationship: part of discovery-benchmark-proteinbench Individual claims | proteinbench primary benchmark evidence Table 4, column(length 100, Quality ppl ↓) Version: 2409.06744v1 | source checked automated source review · 2026-09-18 Audit detailsPrimary-source transcription with no human sign-off and no independent reproduction. Field: Claim: proteinbench-association-seq-length-100-quality-ppl Source artifact SHA-256: Hash scope: Exact retrieved primary paper artifact bytes. |
Sources and history
Release 2026-09-17-134cd1815de8 · Record review: source checked
1 source records and release history
- proteinbench primary benchmark evidence · Original source · 2409.06744v1
Technical metadata and extraction receipts
Stable ID: proteinbench-task-seq-length-100-quality-ppl
- areas
- proteins-complexes
- tasks
- length 100, Quality ppl
- metric
- ppl
- metric direction
- lower
- dataset
- Generated sequences at fixed lengths
- protocol
- Unconditional sequence generation at a fixed length. Values are the mean and standard deviation over repeated runs.
- source locator
- Table 4, column(length 100, Quality ppl ↓)
- comparison panels
- id: proteinbench-panel-seq-length-100-quality-ppl; title: ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl; protocol id: proteinbench-task-seq-length-100-quality-ppl; dataset id: proteinbench-dataset-generated-sequences-at-fixed-lengths; metric: ppl; unit: score; direction: lower; result ids: proteinbench-result-progen-2-700m-seq-length-100-quality-ppl-ppl; proteinbench-result-evodiff-seq-length-100-quality-ppl-ppl; proteinbench-result-dplm-650m-seq-length-100-quality-ppl-ppl; proteinbench-result-esm3-1-4b-seq-length-100-quality-ppl-ppl; source ids: expansion-p3-proteinbench; source locator: Table 4, column(length 100, Quality ppl ↓); context: Every method ProteinBench reports on length 100, Quality ppl, scored with ppl on Generated sequences at fixed lengths.; caveats: Author-reported numbers, source checked but not independently reproduced.; Many of these metrics are better when lower, including scRMSD, perplexity and binding energy. The direction comes from the arrow in the paper's own header.; Rows such as Native PDBs and RAbD are the natural reference the authors include for scale, not a competing method.; review: method: automated_source_review; date: 2026-09-18
Related records
- part of: ProteinBench
- subject: ProteinBench SEQ-LENGTH-100-QUALITY-PPL: part of discovery-benchmark-proteinbench
- benchmark: DPLM (650M) on ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl
- benchmark: ESM3 (1.4B) on ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl
- benchmark: EvoDiff on ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl
- benchmark: Progen 2 (700M) on ProteinBench SEQ-LENGTH-100-QUALITY-PPL: length 100, Quality ppl