Writing study: Exploring Synthesizable Chemical Space with Iterative Pathway Refinements

ICLR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICLR 2026 oral writing corpus

Seul Lee, Karsten Kreis, Srimukh Veccham, Meng Liu, Danny Reidenbach, Saee Paliwal, Weili Nie, Arash Vahdat · Official ICLR 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: Learning & optimization algorithms (primary) · Inference, algorithms & systems efficiency.

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

23 physical pages

Before references / appendix

≈ 9.82 page-equivalents

Main figures / tables

6 / 5

Attached appendix figures / tables

4 / 4

Attached appendix span

≈ 9.9 page-equivalents

Before references / appendix: ≈ 42.7%References / intervening material: ≈ 14.26%Attached appendix: ≈ 43.04%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	≈ 2.14 pp.	9.3%
2 RELATED WORK	3	≈ 0.64 pp.	2.78%
3 BACKGROUND	4	≈ 0.7 pp.	3.04%
4 METHOD	5	≈ 2.34 pp.	10.17%
5 EXPERIMENTS	7	≈ 3.12 pp.	13.57%
6 CONCLUSION	10	≈ 0.18 pp.	0.78%
REFERENCES	10	≈ 3.28 pp.	14.26%
Appendix	14	≈ 0.04 pp.	0.17%
A ITERATIVE REFINEMENT CYCLE OF REASYN	14	≈ 0.52 pp.	2.26%
B OVERVIEW OF EDIT FLOW	14	≈ 0.9 pp.	3.91%
C DETAILS ON PATHWAY REPRESENTATION AND MODEL	15	≈ 1.73 pp.	7.52%
D DETAILS ON TRAINING	17	≈ 0.37 pp.	1.61%
E DETAILS ON INFERENCE	17	≈ 1.93 pp.	8.39%
F ADDITIONAL EXPERIMENTAL RESULTS	19	≈ 4.4 pp.	19.13%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 2
• Figure 3 · p. 3
• Table 1 · p. 7
• Table 2 · p. 7
• Figure 4 · p. 8
• Table 3 · p. 9
• Table 4 · p. 10
• Table 5 · p. 10
• Figure 6 · p. 10
• Figure 5 · p. 10

Attached appendix

• Figure 7 · p. 16
• Table 6 · p. 19
• Table 7 · p. 19
• Table 8 · p. 19
• Table 9 · p. 20
• Figure 8 · p. 21
• Figure 9 · p. 22
• Figure 10 · p. 23

Argument and organization

1. Projection rather than unconstrained generation

   Defines the chemical search space and separates analog generation from fixed-target retrosynthesis.

2. Representation, bidirectional search and holistic edits

   Introduces three complementary operations before testing reconstruction and downstream optimization.

Problem, claim and contribution

The introduction explains why heuristic synthesizability scores are insufficient and why projection can complement an existing molecular optimizer, giving the contribution a practical interface.

Sections 1 and 3, pages 1–4. · Main paper

Argument and section order

Each application subsection separates setup, metric definitions and results; reconstruction, goal-directed search and hit expansion answer different questions.

Sections 5.1, 5.3 and 5.4, pages 7–10. · Main paper

Figures, captions and visual explanation

Tree traversal in Figure 2 precedes the token-level architecture in Figure 3; consistent bottom-up/top-down colors connect structural intuition to implementation.

Figures 2–3, pages 2–3. · Main paper

Tables, comparisons and evidence

Figure 4 plots reconstruction against sampling time rather than only iteration count, exposing test-time compute trade-offs across ablated components.

Figure 4 and Section 5.2, page 8. · Main paper

The appendix discloses that the sEH comparison uses different oracle budgets and top-k selection protocols; this important qualification belongs beside any simplified performance table.

Appendix E.2, page 18; Table 4, page 10. · Main paper

Appendix and reproducibility

Pseudocode, literal pathway examples, vocabulary, model sizes, offline pair generation cost and fallback reaction handling make the method reproducible beyond its schematic.

Appendices A–E, pages 14–19. · Main paper

What to reuse—and what not to infer

Transferable lesson: Explain a structured generative method at three levels—tree operation, sequence representation and executable search—and measure the quality–compute trade-off.

Boundary: Rule-compatible pathways and proxy binding scores are not wet-lab synthesis or biological validation. Compare the two-model total rather than each component's size separately, and retain unequal selection/budget protocols when discussing superiority.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-09-19

Reading status

Full text + selected visuals inspected

Main text

Pages 1–10, Sections 1–6; references on pages 10–13.

Appendix / supplement

Pages 14–23, Appendices A–F, including algorithm, Edit Flow details, representation, training/inference and example pathways.

Visually inspected PDF pages

3, 8

• Official ICLR 2026 proceedings conference PDF; PDF creation metadata: Tue Feb 10 01:44:45 2026 EST · 23 pages