- Rating
- 3
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper proposes PerturbCellRL, an RL-based post-training framework for single-cell perturbation generation. It combines three rewards—Energy Witness, Expression Support, and Pathway Discrimination—to address suboptimal distribution fitting in pretrained Flow Matching models. Experiments demonstrate improved population-level distribution alignment and biological plausibility.
Critical strengths
-
Complementary rewards. The three rewards encourage distributional coverage and reduce overconcentration, statistical expression plausibility, and biological consistency using pathway knowledge.
-
Theoretical support. Energy Witness is theoretically linked to minimizing population-level distributional discrepancy.
-
Experimental validation. Extensive evaluations and ablations on genetic and chemical perturbation datasets support the method's effectiveness.
Critical weaknesses
The motivation regarding cellular response heterogeneity in the Introduction could be misleading and would benefit from clarification.
The authors motivate their work by noting that the same mean expression change may result from weak responses across many cells or strong responses in a small subset. However, this example may lead readers to interpret the method as learning how individual cells respond differently based on their initial states.
During scDFM FM training, control cells and perturbed cells are independently sampled without cell-level correspondence. Consequently, for a given perturbation , any control cell can be paired with samples from the same target distribution , providing no direct supervision for state-dependent transformations. Under this independent-sampling assumption, the theoretical optimum of the standard FM objective can therefore ignore entirely.
Moreover, none of the three rewards directly constrains the relationship between initial cell states and their responses. Therefore, reproducing heterogeneous response patterns does not necessarily imply that the model has learned which initial cells are more likely to respond strongly.
Although the paper acknowledges the unpaired nature of the data, the Introduction could more clearly distinguish heterogeneity in post-perturbation cell populations from initial-state-dependent individual response heterogeneity. Clarifying this distinction would better align the motivation with the method's actual scope.
Questions for authors
Could the authors clarify the motivation for retaining the initial control cell u as a conditioning input, given the unpaired nature of the training data?
In particular, does PerturbCellRL improve the model's ability to capture individual differences in u compared with the pretrained scDFM baseline? An analysis of how different control cells affect the generated responses under the same perturbation would help clarify whether the improvements extend beyond population-level distribution fitting.
3 comments
Response
I appreciate the authors' response, which addresses most of my concerns. However, the new experiments confirm no detectable dependence on individual . While retaining ensures consistency with scDFM, this does not establish its necessity. Its potential benefits in multi-context settings also remain unverified. I believe this limitation should be explicitly acknowledged, as retaining an unvalidated conditioning input warrants stronger justification.
Response to Reviewer LdNh (Part 2)
We thank the reviewer for the follow-up and agree that this should be stated explicitly. PerturbCellRL's contribution lies in the post-training process, not the base model. We therefore keep scDFM's inputs, including , unchanged, so that the gains come from post-training alone. We do not claim that is necessary, nor that post-trained scDFM is the best possible formulation, and whether helps in multi-context settings remains untested.
PerturbCellRL also applies to other single-cell flow-matching models. In the revised PDF, we added a new experiment on a second base model, scLDM (App. G), which has no control-cell input. Post-training improves all nine metrics there as well, so the gains generalize beyond scDFM and do not rely on . Mean std over 3 generation seeds on scLDM's Replogle benchmark:
| Model | Energy | MMD | SW1 | NES-Sp. | MAE | DE-Sp. | DS | ||
|---|---|---|---|---|---|---|---|---|---|
| scLDM | 2.3945 0.0159 | 0.0154 0.0003 | 0.1979 0.0007 | 0.3706 0.0060 | 0.1329 0.0006 | 0.3978 0.0026 | 0.3167 0.0056 | 0.3270 0.0016 | 0.7518 0.0032 |
| + PerturbCellRL | 2.1088 0.0121 | 0.0098 0.0002 | 0.1829 0.0005 | 0.4030 0.0035 | 0.1210 0.0004 | 0.4304 0.0020 | 0.3369 0.0045 | 0.3514 0.0007 | 0.7617 0.0044 |
As suggested, we now state this in the Limitations: "We retain scDFM's control-cell input to post-train it unchanged. Within a context, the flow-matching optimum does not depend on , and whether helps in multi-context settings remains untested. Our second base model, scLDM, has no such input (Appendix G)." Sec. 3.1 now likewise describes only as an input inherited from scDFM.
Response to Reviewer LdNh
We thank the reviewer for the encouraging feedback. We appreciate the reviewer's recognition of the complementary rewards, the theoretical link between the witness and distributional alignment, and the extensive evaluations. As suggested, we rewrote the heterogeneity example, stated the role of u in Sec. 3.1, and analyzed u-dependence empirically (Q3), which we have also added to the paper as the new App. I.
Q1. Population-level vs. initial-state-dependent heterogeneity.
We agree that the distinction should be explicit. Our prediction target is the distribution of the perturbed population Pc, not how each initial state u responds. The example contrasts two perturbed populations with the same mean. These differ in distribution, so they can be distinguished from unpaired data: Energy and MMD separate them, whereas mean-based metrics cannot. Accordingly, our objective (Eq. 3) and all metrics are defined on the generated population Qθ,c. As the reviewer notes, neither the FM loss nor our rewards constrain the map from u to its response, and none of our claims relies on it. The revised Introduction now reads: "…the same shift in average expression could come from a perturbed population in which all cells shift slightly, or one in which a small subset shifts strongly. […] the data reveal how the perturbed population is distributed, but not how each individual cell would have responded."
Q2. Why retain u as an input?
We retain u for two reasons:
In scDFM, the flow starts from Gaussian noise and u enters only as a conditioning input; it does not map each control cell to a perturbed counterpart. Since u and x1 are sampled independently given c, the flow-matching optimum v⋆=E[x1−x0∣xt,t,c] does not depend on u within a context, as the reviewer notes. The revised Sec. 3.1 now states this role of u.
Q3. Does post-training change dependence on individual u?
No. Neither scDFM nor PerturbCellRL depends detectably on individual differences in u, as expected from the flow-matching optimum.
We test whether each generated cell correlates more with its own control cell u than with other control cells, as it would if the output depended on u. For each held-out condition, each model generates 128 cells, one from each of 128 control cells, with the same control cells and noise for both models. After subtracting the condition mean, we average the correlation between each generated cell and its own control cell, and compare it with random pairings (permutation test, 199 permutations). We ran this on all 308 held-out conditions, counting each Norman fold separately. The test is sensitive: when we add an artificial u-dependence to scDFM's outputs, it detects it in every condition once that dependence explains about 0.5% of the variation among generated cells. Since this correlation captures only dependence along the same genes, we also run a distance-correlation test between the control cells and the generated cells (top 10 principal components of each, 199 permutations), which also captures nonlinear and cross-gene dependence.
Both models are at chance level in both tests: about 5% of conditions would reach p<0.05 if the output ignored u. The variation among generated cells therefore comes from the noise draw, and the gains of PerturbCellRL are at the population level. Learning which control states respond more strongly would require paired or lineage-resolved measurements, which these screens do not provide; we see this as an interesting direction for future work.
We hope these clarifications address the reviewer's concerns and would be glad to answer any further questions.