Writing study: Protein Autoregressive Modeling via Multiscale Structure Generation

ICML overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICML 2026 oral writing corpus

Yanru Qu, Cheng-Yen Hsieh, Zaixiang Zheng, Ge Liu, Quanquan Gu · Official ICML 2026 oral record

Observed writing choices, not acceptance rules. This study examines the inspected manuscript version. It does not establish why the paper was accepted or that its presentation choices should be copied unchanged.

Figure, table and section ledger

Editorial contribution type: Architecture & representation (primary).

Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.

Canonical PDF

21 physical pages

Before references / appendix

≈ 8.81 page-equivalents

Main figures / tables

6 / 5

Attached appendix figures / tables

4 / 10

Attached appendix span

≈ 8.91 page-equivalents

Before references / appendix: ≈ 41.95%References / intervening material: ≈ 15.62%Attached appendix: ≈ 42.43%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Section-by-section lengths and source pages

The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.

Top-level heading	Starts on PDF page	Estimated span	Share of entire PDF
1. Introduction	1	≈ 1.46 pp.	6.95%
2. Background and Related Work	3	≈ 0.54 pp.	2.57%
3. Protein Autoregressive Modeling	3	≈ 2.3 pp.	10.95%
4. Experiments	5	≈ 3.39 pp.	16.14%
5. Discussion	9	≈ 0.34 pp.	1.62%
Acknowledgments	9	≈ 0.05 pp.	0.24%
Impact Statement	9	≈ 0.12 pp.	0.57%
References	9	≈ 3.27 pp.	15.57%
A. Implementation and Evaluation Details	13	≈ 1.72 pp.	8.19%
B. Datasets	14	≈ 0.45 pp.	2.14%
C. More Empirical Analysis	15	≈ 5.21 pp.	24.81%
D. Other Related Work	20	≈ 0.23 pp.	1.1%
E. The Use of Large Language Models	20	≈ 0.08 pp.	0.38%
F. Discussion on Future Work	20	≈ 1.23 pp.	5.86%

All detected figure and table locations

Main text

• Figure 1 · p. 2
• Figure 2 · p. 2
• Table 1 · p. 5
• Figure 3 · p. 6
• Figure 4 · p. 7
• Figure 5 · p. 7
• Table 2 · p. 8
• Table 3 · p. 8
• Figure 6 · p. 8
• Table 4 · p. 8
• Table 5 · p. 9

Attached appendix

• Table 6 · p. 13
• Table 7 · p. 15
• Figure 7 · p. 15
• Table 8 · p. 15
• Figure 8 · p. 16
• Figure 9 · p. 16
• Table 9 · p. 17
• Table 10 · p. 17
• Table 11 · p. 18
• Table 12 · p. 18
• Table 13 · p. 18
• Table 14 · p. 19
• Figure 10 · p. 20
• Table 15 · p. 20

Argument and organization

1. Problem and contribution

   Defines limitations of discretization and residue-wise ordering before proposing a coarse-to-fine order that preserves within-scale structural interactions.

2. Argument sequence

   Multiscale representation comes before the autoregressive factorization and decoder objective; exposure bias is then addressed before evaluation.

3. Evidence and verification

   Sampling analysis contrasts mixed SDE/ODE execution with both aggressive SDE step reduction and all-ODE alternatives, tying the speed claim to a designability control.

Problem, claim and contribution

Defines limitations of discretization and residue-wise ordering before proposing a coarse-to-fine order that preserves within-scale structural interactions.

Introduction and Section 3, pp. 1-4 · Main paper

Argument and section order

Multiscale representation comes before the autoregressive factorization and decoder objective; exposure bias is then addressed before evaluation.

Sections 3.1-3.3, pp. 4-5 · Main paper

Figures, captions and visual explanation

The method diagram and generated-scale grid share a coarse-to-fine reading direction. The motif figure uses a stable motif color across input/generated rows, while scaling plots retain metric direction arrows.

Figures 1-2, p. 2; Figures 4-5, p. 7; both PNGs inspected · Main paper

Tables, comparisons and evidence

Sampling analysis contrasts mixed SDE/ODE execution with both aggressive SDE step reduction and all-ODE alternatives, tying the speed claim to a designability control.

Section 4.3, pp. 7-8; Appendix C.1 and Table 8, p. 15 · Main paper

Appendix and reproducibility

Runtime tables state model sizes, conditioning, lengths, sample counts and steps; later subsections distinguish completed experiments from proposed multi-chain/molecular-dynamics extensions.

Table 8, p. 15; Appendix F, pp. 20-21 · Main paper

Clarity and writing conventions

A sculpting analogy introduces the scale order, followed immediately by explicit downsampling and conditional-probability definitions.

Introduction, p. 2; Sections 3.1-3.2, p. 4 · Main paper

What to reuse—and what not to infer

Transferable lesson: Reuse one representation hierarchy across the overview, formal factorization and qualitative output sequence.

Boundary: Plausible computational scaffolds and self-consistency metrics are not wet-lab validation; future biomolecular extensions are not demonstrated capabilities.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-10-07

Reading status

Writing-focused manuscript scan + selected visuals inspected

Main text

Writing-focused distributed manuscript scan, not exhaustive technical/proof validation. Sections 1-5, pp. 1-9: autoregressive-order motivation, multiscale construction, exposure-bias mitigation, generation/generalization and sampling analysis.

Appendix / supplement

Writing-focused appendix scan; code, data and external assets not executed. Appendices A-F, pp. 13-21: implementation, metrics/data, extended ablations, sampling/runtime comparisons, motif benchmarks, attention views and future directions.

Visually inspected PDF pages

2, 7

This is a writing-focused scan of the manuscript and the stated supplementary scope: framing, organization, visual explanation, evidence presentation and wording. It is not a line-by-line proof audit, reproduction of results or exhaustive scientific review.

• Published ICML 2026 proceedings PDF, PMLR volume 306; retrieved 2026-10-07. Public OpenReview decision notes were not independently reverified. · 21 pages