- Rating
- 1
- Confidence
- 3
- Informativeness
- 1
- Presentation
- 2
- Soundness
- 2(+1)
Read review
Paper summary
This paper introduces TFPO, which applies GRPO to steer flexible-size molecular generation through a parameterized-action formulation. The method is applied to Morph, and the experiments are conducted on QM9, GEOM-Drugs, and CrossDocked2020, mainly evaluated in terms of property scores, structural quality, and diversity.
Critical strengths
The paper is generally well written and organized. The presentation is relatively clear, and the visualizations provide useful support for understanding the proposed framework and experimental results.
Critical weaknesses
-
The novelty of this study may be limited. The work does not appear to introduce new algorithms, theories, frameworks, or insights. Instead, this work applies existing standard parameterized-action MDP and GRPO to Morph, which may be methodologically incremental. Additionally, although this is an application-driven study, it does not appear to clearly propose a new application-driven method or direction to address a specific challenging question in the field and demonstrate its necessity. Thus, the new knowledge identified from this work that can contribute to the community may be limited.
-
Several major concerns may be raised according to the experimental evidence provided. (1) There are no key ablations to show trans-dimensional policy formulation is necessary, adaptive cardinality is effective (molecular size is not the only variable), and fixed-variance Gaussian noise design is reasonable and robust. (2) OOD experiments on QM9 may imply severe 3D generation failure. Only ~2.8% of the raw generated molecules pass all checks, while the reported result of 65% is after MMFF force-field minimization. This may mean the claimed OOD generation ability is supported only after external force-field repair and the generation of molecules with large size may not be reliable. Additionally, the reward for QM9 experiments appears to be based only on atom count. This may just encourage the model to generate larger molecules rather than OOD discovery. (3) Comparing optimization-based frameworks with amortized generative models (e.g., DiffSBDD, MolPilot) may be unfair. For the relatively fair baseline comparison (DePPA vs. the proposed method), the key metric (vina score) of the proposed method is uncompetitive. In addition, the diversity in the table is also reduced a lot and not competitive (DePPA vs. Morph vs. the proposed method: 0.78 vs. 0.58 vs. 0.41), implying potential mode collapse. Given this, it may be unclear if the proposed method is more worthwhile than the existing methods, since the computational cost of the proposed method may also be relatively large.
Questions for authors
-
What are the substantial contributions or novelties of this work?
-
Can the authors provide more convincing evidence to show the effectiveness and necessity of the proposed method?
8 comments
Response to Weakness 2 - Structural ablations and adaptive cardinality
there are no key ablations to show trans-dimensional policy formulation is necessary, adaptive cardinality is effective (molecular size is not the only variable)
We agree with the reviewer that the contribution of structural optimization should be isolated experimentally. We therefore conduct controlled ablations on CrossDocked2020 and QM9, testing both implicit size adaptation and explicit cardinality extrapolation.
CrossDocked2020: We compare full TFPO against ablation A2 (reference structural rates), where insertion/deletion rates are taken from the frozen pretrained model. Experiments use the same backbone, reward, and budget across 100 pockets and 3 training seeds, with a new backbone trained for 100 epochs.
| Method | VS Avg ↓ | VM Avg ↓ | VD Avg ↓ | QED ↑ | SA ↑ | Div ↑ | Size | Complete ↑ | PB-valid ↑ |
|---|---|---|---|---|---|---|---|---|---|
| Reference (published) | -6.36 | -6.71 | -7.45 | 0.48 | 0.73 | — | 22.8 | 100.0% | — |
| Morph base (no RL) | -5.72 | -6.83 | -7.97 | 0.55 | 0.71 | 0.73 | 22.8 | 97.4% | 0.81 |
| TFPO (ours) | -7.14 ± 0.02 | -8.07 ± 0.03 | -8.98 ± 0.03 | 0.61 ± 0.001 | 0.80 ± 0.003 | 0.75 ± 0.003 | 21.3 ± 0.2 | 97.7% ± 0.0% | 0.85 ± 0.003 |
| TFPO, A2 ref-struct | -6.79 ± 0.08 | -7.72 ± 0.08 | -8.67 ± 0.06 | 0.61 ± 0.004 | 0.80 ± 0.002 | 0.76 ± 0.003 | 20.6 ± 0.2 | 97.7% ± 0.2% | 0.85 ± 0.004 |
On CrossDocked, where size adapts indirectly to the downstream objective, full TFPO improves Vina Score from −5.72 (base) to −7.14 ± 0.02 kcal/mol. Using reference insertion/deletion rates (A2) yields −6.79 ± 0.08. Thus, learned structural adaptation yields a 0.35 kcal/mol advantage over A2, approximately 25% of the total improvement, demonstrating that structural optimization contributes to binding affinity even without an explicit cardinality reward.
QM9 - Cardinality extrapolation: We additionally evaluate two matched Morph ablations: A1 removes structural log-probabilities from the policy gradient, while A2 uses frozen reference insertion/deletion rates. We also compare against fixed-cardinality SemlaFlow with an externally optimized size distribution (Thompson-sampling bandit) based on Wang et al. (2026) [1]. All use the same reward and RL budget, with 3 seeds and 3 curriculum rounds of 600 steps.
| Method | R1 atoms | R2 atoms | R3 atoms | R3 Max | R3 Validity | R3 PoseBusters |
|---|---|---|---|---|---|---|
| Full TFPO | 44.4 ± 2.7 | 59.5 ± 4.7 | 77.7 ± 3.8 | 86 | 0.866 ± 0.035 | 0.823 ± 0.049 |
| A1: stop-grad | 30.5 ± 0.1 | 34.4 ± 3.3 | 36.8 ± 4.3 | 50 | 0.976 ± 0.004 | 0.974 ± 0.005 |
| A2: ref rates | 28.5 ± 0.2 | 30.3 ± 1.6 | 31.5 ± 1.8 | 41 | 0.952 ± 0.018 | 0.947 ± 0.018 |
| SemlaFlow + bandit | 28.5 ± 2.8 | 14.9 ± 1.7 | 13.4 ± 2.4 | 21 | 0.001 ± 0.001 | 0.000 |
The gap between the full TFPO and both ablations increases over the curriculum rounds. Full TFPO achieves +60.1 atoms over the pretrained mean, compared with +19.2 (A1) and +13.9 (A2). The externally controlled SemlaFlow baseline initially reaches 28.5 atoms but collapses to 0.001 validity by round 3, whereas TFPO reaches 77.7 atoms at 0.866 validity.
Together, these ablations demonstrate that directly optimizing structural decisions benefits both cardinality extrapolation (QM9) and downstream binding affinity without an explicit size reward (CrossDocked).
[1] Wang, K., Janet, J. P., & Tibo, A. (2026). Controllable molecular generation with fine-tuned flow-matching model. Communications Chemistry, 9, 291. https://doi.org/10.1038/s42004-026-02188-z
and fixed-variance Gaussian noise design is reasonable and robust.
We agree with the reviewer that the robustness and design of the fixed variance gaussian noise is not ablated. The design of the gaussian kernel follows the RL for flow policies literature (Liu et al., 2025; Zhang et al., 2025) to give a tractable transition density to densities.
We evaluated different noise scales, linear decay, and learned noise (initialized at σ=0.05) on QM9 curriculum round 3, over 3 training seeds:
| σ explore | Schedule | Mean atoms | Max | Validity | Diversity | PoseBusters |
|---|---|---|---|---|---|---|
| 0.01 | Fixed | 68.2 ± 3.0 | 77 | 0.896 ± 0.014 | 0.723 ± 0.006 | 0.795 ± 0.102 |
| 0.05 (ours) | Fixed | 75.3 ± 4.3 | 86 | 0.867 ± 0.031 | 0.722 ± 0.008 | 0.835 ± 0.066 |
| 0.10 | Fixed | 72.1 ± 5.1 | 83 | 0.903 ± 0.046 | 0.720 ± 0.004 | 0.896 ± 0.040 |
| 0.05 | Linear decay | 70.2 ± 5.0 | 80 | 0.891 ± 0.027 | 0.731 ± 0.009 | 0.854 ± 0.068 |
| Learned | — | 76.7 ± 4.0 | 86 | 0.885 ± 0.049 | 0.716 ± 0.008 | 0.814 ± 0.107 |
Performance of the gaussian noise remains robust across the tested configurations. Larger fixed noise improves geometric validity, while learned noise achieves comparable cardinality extrapolation, suggesting that fixed variance is a reasonable but not uniquely optimal choice.
Response to QM9 OOD validity and reward formulation
OOD experiments on QM9 may imply severe 3D generation failure. Only ~2.8% of the raw generated molecules pass all checks, while the reported result of 65% is after MMFF force-field minimization. This may mean the claimed OOD generation ability is supported only after external force-field repair and the generation of molecules with large size may not be reliable.
We agree with the reviewer that the submitted QM9 results show raw geometric degradation, and attribute the gap to an undertrained backbone and to reward mis-specification rather than the method.
We reran the experiment with a longer-trained backbone, evaluating 3 independent training seeds (3 × 1000 molecules per seed) without any force-field relaxation. The same TFPO procedure reaches 77.7 ± 3.8 atoms (4.4× the training mean, max 86), with RDKit validity of 0.866 ± 0.035 and raw PoseBusters validity of 0.823 ± 0.049, compared with 0.028 in the submitted runs. Conditional on chemical validity, PoseBusters validity reaches 0.950. Thus, the improved extrapolation does not depend on external geometry repair.
We would add that the QM9 reward contains no geometric term and the gate is RDKit validity, which stays flat while size grows. Geometry was never rewarded, so its drift reflects an unconstrained quantity rather than an inability to generate reliably at large size. In some other cases where a terminal reward does impact geometry like on GEOM-Drugs (Section 3.2) PoseBusters increases from 0.878 to 0.943 while PoseCheck strain falls.
Additionally, the reward for QM9 experiments appears to be based only on atom count. This may just encourage the model to generate larger molecules rather than OOD discovery.
We agree with the reviewer that the QM9 reward is simply size-extrapolation. It was built as such, to test the trans-dimensional machinery without any other property confounds. The property-driven claims rest on the two other experiments which both don’t directly reference size, showcasing validity increase at a preserved size distribution on GEOM-Drugs and a five term binding reward on CrossDocked under which size adapts bidirectionally.
Response to Weakness 1 and Q1 - Novelty and contributions
We thank the reviewer for raising this important point and agree that the methodological contribution of TFPO should be more clearly articulated.
Our contribution is a general formulation connecting trans-dimensional generative modeling with parametrized-action reinforcement learning, together with a concrete realization that enables reward driven optimization of a dimension changing generative process.
We acknowledge the underlying RL principles are established, the new contribution lies in identifying and operationalizing their connection with trans-dimensional geometric generators, deriving the required likelihood for a non-trivial jump-flow process and establishing the value of optimizing structural decisions directly. We have strengthened this contribution presentation in the paper.