- Rating
- 3(+1)
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper proposes ABSORB, which formulates the imputation task as a "selective missing-data" problem and performs selective imputation of scRNA-seq data using an absorbing-state discrete diffusion framework. The paper quantifies numerical recovery at held-out positions with known ground truth, and compares against nine baselines on simulated datasets at four different dropout rates as well as multiple real datasets. The results show that ABSORB has a clear advantage on simulated data, and this advantage widens further as the dropout rate increases.
Critical strengths
This paper frames scRNA-seq imputation as a "selective missing-data" problem and, based on absorbing-state discrete diffusion, realizes selective imputation that "keeps observed entries fixed and reconstructs only candidate dropouts." The problem formulation is clear and clearly distinguished from the continuous-diffusion denoising route. Under a controlled held-out evaluation, the method performs well on multiple recovery metrics.
Critical weaknesses
(1) The paper does not experimentally decompose the contribution of each component, nor does it try alternative decoding architectures (e.g., a conditional autoencoder or an MLP decoder). It is therefore difficult to judge whether the performance comes primarily from absorbing-state diffusion itself or from masked token prediction, the Transformer, tokenization, or confidence-first decoding; although the supplementary material provides a theoretical analysis of the role/timestep embeddings, this is not sufficient to replace an experimental ablation.
(2) ABSORB imputes only candidate dropouts with a mask value of 0, and does not handle "non-zero but unreliable" observations whose counts are deflated by dropout; at the evaluation level, it scores only at held-out positions with known ground truth, whereas most non-diffusion baselines impute the full matrix (scIDPMs and scImpute are themselves selective imputers). Under this setting, imputation performance depends on the reliability of the candidate-dropout mask, yet the current held-out evaluation is not sufficient to comprehensively measure the consequences of mask misclassification—especially the cost of erroneously imputing biological zeros. Its overall imputation utility in real scenarios therefore still requires further validation.
Questions for authors
(1) The current method determines whether a gene is expressed in a cluster solely based on whether that gene appears non-zero at least once in the cluster: once a non-zero appears within the cluster, its zeros are treated as candidate dropouts. This criterion is simple and easily confounded by sequencing contamination, ambient RNA, doublets, batch effects, and the like. It may both misclassify a contaminated biological zero as a dropout (false positive) and retain a genuine dropout that is zero throughout the cluster as a biological zero (false negative). The reasonableness of this assumption deserves further discussion.
(2) The paper mainly constructs dropout through simulated data or by down sampling real data; it remains unclear whether, when applied to real datasets without artificially simulated dropout, the method can recover genuine biological information (such as cell-type marker genes, cell-type proportions, and gene correlations)
10 comments
Evidence of ablation study on ABSORB vs. Masked-MLP Transformer
Matched ablation: ABSORB vs. masked-Transformer controls trained at fixed mask rates of 0.50 and 0.30 (columns .50 and .30), with the same backbone and tokenization. Held-out recovery; bold marks the best of the three.
| Dataset | ABSORB / .50 / .30 | RMSE ABSORB / .50 / .30 | ED ABSORB / .50 / .30 | MMD ABSORB / .50 / .30 | SW ABSORB / .50 / .30 |
|---|---|---|---|---|---|
| Sim D1 | 0.922 / 0.907 / 0.908 | 0.609 / 0.675 / 0.667 | 362 / 449 / 438 | 0.144 / 0.125 / 0.111 | 29.7 / 32.7 / 30.5 |
| Sim D2 | 0.928 / 0.904 / 0.910 | 0.595 / 0.697 / 0.678 | 268 / 305 / 297 | 0.110 / 0.0997 / 0.0927 | 26.4 / 26.5 / 26.7 |
| Sim D3 | 0.918 / 0.906 / 0.910 | 0.634 / 0.670 / 0.652 | 220 / 274 / 281 | 0.0826 / 0.0840 / 0.0845 | 23.5 / 25.2 / 25.0 |
| Sim D4 | 0.942 / 0.917 / 0.922 | 0.549 / 0.670 / 0.644 | 153 / 196 / 241 | 0.0552 / 0.0568 / 0.0653 | 20.0 / 22.3 / 24.7 |
| PBMC68k MCAR50 | 0.197 / 0.209 / 0.214 | 0.478 / 0.485 / 0.496 | 0.125 / 0.117 / 0.113 | 0.00260 / 0.00199 / 0.00190 | 0.0727 / 0.0723 / 0.0671 |
| PBMC68k MNAR25 | 0.334 / 0.244 / 0.234 | 0.412 / 0.424 / 0.484 | 0.0517 / 0.0941 / 0.176 | 0.000705 / 0.00169 / 0.00370 | 0.0487 / 0.0640 / 0.0931 |
| Bone Marrow MCAR50 | 0.392 / 0.324 / 0.310 | 0.703 / 0.834 / 0.787 | 3.37 / 9.45 / 3.90 | 0.0209 / 0.0646 / 0.0272 | 0.570 / 0.879 / 0.609 |
| Bone Marrow MNAR25 | 0.402 / 0.335 / 0.307 | 0.634 / 0.806 / 0.762 | 0.541 / 2.02 / 1.25 | 0.00322 / 0.0122 / 0.00756 | 0.224 / 0.450 / 0.365 |
| Human Brain MCAR50 | 0.280 / 0.239 / 0.232 | 2.543 / 2.638 / 2.731 | 471 / 506 / 536 | 0.0167 / 0.0171 / 0.0175 | 107 / 109 / 110 |
| Human Brain MNAR25 | 0.249 / 0.227 / 0.223 | 2.547 / 2.559 / 2.722 | 95.5 / 121 / 250 | 0.00243 / 0.00311 / 0.00720 | 43.5 / 53.3 / 78.5 |
| VISp MCAR50 | 0.510 / 0.350 / 0.297 | 2.376 / 2.981 / 2.928 | 1,396 / 2,749 / 3,678 | 0.0147 / 0.0269 / 0.0322 | 153 / 211 / 236 |
| VISp MNAR25 | 0.514 / 0.253 / 0.404 | 2.271 / 2.979 / 2.564 | 106 / 535 / 156 | 0.000800 / 0.00451 / 0.00129 | 44.6 / 91.3 / 48.4 |
| Dentate Gyrus MCAR50 | 0.481 / 0.464 / 0.477 | 0.709 / 0.949 / 1.090 | 12.8 / 69.3 / 141 | 0.0397 / 0.162 / 0.260 | 0.736 / 2.03 / 3.52 |
| Dentate Gyrus MNAR25 | 0.438 / 0.426 / 0.433 | 0.768 / 1.140 / 0.798 | 3.73 / 36.8 / 7.41 | 0.0116 / 0.104 / 0.0236 | 0.421 / 1.67 / 0.612 |
| ABSORB better | 13/14 | 14/14 | 13/14 | 11/14 | 13/14 |
Clarifications on candidate-mask reliability and biological zeros
Candidate-mask reliability and biological zeros…
We thank the reviewer for highlighting this concern. We agree that identifying potential dropouts based on the presence of at least one non-zero value within a cluster can be sensitive to noise and may lead to incorrect imputation of biological zeros. In the original submission, we reported results at α = 0, as this operating point prioritizes maximum dropout recovery. However, the selectivity parameter α allows us to control this trade-off by defining a dataset-adaptive expression-frequency threshold based on the α-quantile of expression frequencies associated with observed non-zero entries. Candidate imputations for genes expressed below this threshold within their cluster are retained as zeros, thereby reducing the influence of rare, potentially noisy expression observations. As shown in our sensitivity analysis, α = 0.05 substantially reduces erroneous biological-zero imputation while maintaining high genuine-dropout recovery. Nevertheless, we acknowledge that distinguishing biological zeros from technical dropouts remains challenging, particularly under contamination or severe sparsity, and have included this analysis and the associated limitations in the Appendix section 2.6.
ABSORB at α = 0 is the original submission; the zero-decision rows use α = 0.02 and α = 0.05.
| Method | Biological zeros imputed | Real dropouts recovered |
|---|---|---|
| ABSORB, α = 0 (original submission) | 65.5% | 91.7% |
| ABSORB + zero decision, α = 0.02 | 35.9% | 91.1% |
| ABSORB + zero decision, α = 0.05 | 12.6% | 90.2% |
| SAVER | 0.1% | 26.5% |
| ALRA | 2.4% | 72.6% |
| MAGIC | 3.4% | 84.6% |
| scImpute | 3.5% | 81.4% |
| DCA | 21.6% | 83.7% |
| countsdiff | 32.8% | 65.4% |
| DeepImpute | 91.3% | 94.3% |
Held-out recovery before → after the zero decision (α = 0 → α = 0.05) on the real datasets. Bold marks the better value; unbolded pairs are ties at the reported precision.
| Dataset | RMSE | ED | MMD | SW | |
|---|---|---|---|---|---|
| PBMC68k MCAR50 | 0.197 → 0.123 | 0.478 → 0.487 | 0.125 → 0.178 | 0.00260 → 0.00415 | 0.0727 → 0.0868 |
| PBMC68k MNAR25 | 0.334 → 0.282 | 0.412 → 0.423 | 0.0517 → 0.0514 | 0.000705 → 0.000671 | 0.0487 → 0.0485 |
| Human Brain MCAR50 | 0.280 → 0.298 | 2.543 → 2.568 | 471 → 467 | 0.0167 → 0.0166 | 107 → 107 |
| Human Brain MNAR25 | 0.249 → 0.278 | 2.547 → 2.562 | 95.5 → 93.1 | 0.00243 → 0.00235 | 43.5 → 43.1 |
| Bone Marrow MCAR50 | 0.392 → 0.415 | 0.703 → 0.707 | 3.37 → 3.29 | 0.0209 → 0.0205 | 0.570 → 0.565 |
| Bone Marrow MNAR25 | 0.402 → 0.428 | 0.634 → 0.638 | 0.541 → 0.506 | 0.00322 → 0.00301 | 0.224 → 0.214 |
| VISp MCAR50 | 0.510 → 0.534 | 2.376 → 2.385 | 1,396 → 1,395 | 0.0147 → 0.0146 | 153 → 153 |
| VISp MNAR25 | 0.514 → 0.541 | 2.271 → 2.305 | 106 → 103 | 0.000800 → 0.000772 | 44.6 → 44.2 |
| Dentate Gyrus MCAR50 | 0.481 → 0.500 | 0.709 → 0.714 | 12.8 → 12.6 | 0.0397 → 0.0390 | 0.736 → 0.727 |
| Dentate Gyrus MNAR25 | 0.438 → 0.460 | 0.768 → 0.769 | 3.73 → 3.59 | 0.0116 → 0.0111 | 0.421 → 0.410 |
| Zero decision better | 8/10 | 0/10 | 9/10 | 9/10 | 7/10 (2 ties) |
Clarifications on component contribution
We thank the reviewer for the constructive comments and for recognizing our contribution along with the distinction between selective missing-data reconstruction and continuous denoising.
Component contribution…
We thank the reviewer for raising this important point. To address this we have added a matched masked-Transformer ablation using the same backbone and tokenization. ABSORB performs better in 64/70 dataset–metric comparisons, including 14/14 RMSE comparisons, providing empirical evidence that the gains are not explained by tokenization and masked Transformer prediction alone. We have added this ablation study to the Appendix section 2.9.
Our novelty lies in the combination of the following design choices:
Together, these choices distinguish ABSORB from Gaussian diffusion approaches such as scIDPMs and count-space diffusion with binomial thinning in methods such as CountsDiff, against which ABSORB also demonstrates stronger recovery performance.