q-bio.QMAug 21, 2026
SavePerturbRx: Learning Treatment-Conditioned Latent Transitions for Patient Drug Response Prediction
Organizations: Department of Computer Science, University of Minnesota, Minneapolis, MN, USA · Computational Biology Branch, National Library of Medicine, Bethesda, MD, USA · Therapeutic Development Branch, National Library of Medicine, Bethesda, MD, USA · Department of Electrical and Electronic Engineering, Inha University, Incheon, South Korea · Department of Electrical Engineering and Computer Science, University of Michigan, Ann Arbor, MI, USA
Abstract
Scarce data and tumor heterogeneity limit patient-level cancer treatment-response prediction. Existing approaches predict response from pretreatment molecular profiles and drug representations, without explicitly modeling the molecular changes expected under treatment. We propose PerturbRx, a treatment-conditioned representation learning framework that learns intervention-induced latent transitions and uses them as patient-drug response features. PerturbRx trains a drug- and dose-conditioned transition predictor from context-matched but cell-unpaired control and treated single-cell populations, then freezes and transfers the predictor to pretreatment patient profiles without requiring post-treatment measurements. The transition is combined with patient and drug representations to predict response. Across TCGA and patient-derived xenograft benchmarks, PerturbRx achieves the strongest aggregate predictive performance among the evaluated methods. These results support perturbation-pretrained latent transitions as useful representations for patient-level drug-response prediction.
Figures & tables
Figure 1: Overview of population-level PerturbRx. (A) Source-stage learning from context-matched but unpaired control and treated single-cell populations. A drug- and dose-conditioned predictor maps control embeddings and intervention features to cell-wise transitions, which are added to the controls to form post-intervention embeddings. The predictor is trained using distributional, transition-direction, transition-displacement, and transition-magnitude objectives. (B) Transfer of the transition predictor to patient drug-response prediction. The predictor generates an intervention-conditioned transition , and the response head predicts response from .
Figure 2: Source-stage perturbation prediction. A. Observed untreated-to-treated shift for an example held-out condition. B. Alignment between predicted and observed treated states for the same condition. Crosses denote population means. C. Performance across 800 randomly held-out conditions. Bold , underline , and † indicate the best, second-best, and third-best results, respectively.
| Overall | Cisplatin | Fluorouracil ‡ | Gemcitabine ‡ | Sorafenib | Temozolomide | |||||||
| Methods | AUROC | AUPRC | AUROC | AUPRC | AUROC | AUPRC | AUROC | AUPRC | AUROC | AUPRC | AUROC | AUPRC |
| CODE-AE | 0.534 † 0.053 | 0.564 0.051 | 0.609 † 0.073 | 0.666 † 0.058 | 0.548 † 0.147 | 0.516 0.134 | 0.360 0.023 | 0.413 0.023 | 0.631 † 0.015 | 0.689 † 0.040 | 0.581 † 0.028 | 0.587 † 0.031 |
| DeepSADR | 0.530 0.011 | 0.519 0.027 | 0.569 0.103 | 0.596 0.093 | 0.749 0.113 | 0.768 0.137 | 0.573 † 0.122 | 0.601 0.121 | 0.720 0.126 | 0.712 0.148 | 0.508 0.111 | 0.528 0.106 |
| TransDRP | 0.494 0.061 | 0.500 0.068 | 0.484 0.176 | 0.598 0.104 | 0.354 0.094 | 0.391 0.044 | 0.593 0.068 | 0.569 0.070 | 0.473 0.220 | 0.581 0.156 | 0.496 0.217 | 0.552 0.175 |
| WISER | 0.580 0.053 | 0.563 † 0.062 | 0.709 0.151 | 0.756 0.141 | 0.480 0.168 | 0.526 † 0.175 | 0.578 0.076 | 0.547 † 0.081 | 0.467 0.105 | 0.574 0.115 | 0.589 0.124 | 0.621 0.118 |
| PerturbRx | 0.626 0.019 | 0.600 0.029 | 0.684 0.056 | 0.733 0.093 | 0.674 0.104 | 0.645 0.134 | 0.551 0.082 | 0.530 0.055 | 0.753 0.213 | 0.801 0.174 | 0.653 0.142 | 0.680 0.107 |
Table 1: Patient-response prediction on the TCGA-186 benchmark.
| Input representation / control | AUROC | AUPRC |
| Patient embedding | 0.547 0.040 | 0.518 0.062 |
| Patient + drug embedding | 0.581 0.019 | 0.556 0.055 |
| Patient + drug + shuffled transition | 0.566 0.029 | 0.551 0.048 |
| Patient + drug + random transition | 0.589 † 0.021 | 0.574 † 0.039 |
| Patient + drug + predicted post-state | 0.609 0.051 | 0.582 0.062 |
| PerturbRx (Patient + drug + transition) | 0.626 0.019 | 0.600 0.029 |
Table 2: Ablation results on TCGA-186.
| Overall | Cisplatin | Docetaxel ‡ | Fluorouracil ‡ | Gemcitabine ‡ | Paclitaxel ‡ | |||||||
| Methods | AUROC | AUPRC | AUROC | AUPRC | AUROC | AUPRC | AUROC | AUPRC | AUROC | AUPRC | AUROC | AUPRC |
| CODE-AE | 0.525 † 0.061 | 0.649 0.193 | 0.600 † 0.028 | 0.841 † 0.027 | 0.453 0.056 | 0.557 0.038 | 0.545 0.024 | 0.683 0.042 | 0.472 † 0.044 | 0.366 0.043 | 0.556 0.064 | 0.799 0.040 |
| DeepSADR | 0.565 0.147 | 0.735 0.093 | 0.511 0.071 | 0.807 0.121 | 0.724 0.244 | 0.812 0.160 | 0.565 0.103 | 0.741 0.072 | 0.675 0.055 | 0.582 0.234 | 0.351 0.093 | 0.732 † 0.258 |
| TransDRP | 0.508 0.089 | 0.665 † 0.174 | 0.613 0.164 | 0.885 0.054 | 0.425 0.241 | 0.588 0.197 | 0.587 0.192 | 0.710 † 0.178 | 0.422 0.083 | 0.418 † 0.079 | 0.493 0.253 | 0.726 0.122 |
| WISER | 0.507 0.016 | 0.650 0.151 | 0.480 0.040 | 0.829 0.017 | 0.515 † 0.074 | 0.607 † 0.055 | 0.511 0.032 | 0.667 0.034 | 0.512 0.032 | 0.422 0.031 | 0.519 † 0.051 | 0.724 0.033 |
| PerturbRx | 0.692 0.055 | 0.787 0.055 | 0.631 0.168 | 0.864 0.085 | 0.711 0.141 | 0.802 0.112 | 0.562 † 0.116 | 0.748 0.098 | 0.472 † 0.095 | 0.395 0.052 | 0.715 0.135 | 0.842 0.081 |
Table 3: Patient-response prediction on the TCGA-508 patient-grouped benchmark.
Figure 3: Predicted transition geometry in TCGA-508. PCA was fit jointly to pretreatment and predicted post-intervention embeddings for all 508 treatment episodes. Contours show the two state distributions within each response group, and black arrows connect their group centroids. Panel annotations report sample size, mean predicted transition magnitude , and mean episode-level displacement in the two-dimensional PCA projection. The reported 2D shift is the mean paired episode-level displacement, not the centroid-arrow length.
| Input representation / control | AUROC | AUPRC |
| Patient embedding | 0.638 0.038 | 0.756 0.041 |
| Patient + drug embedding | 0.681 † 0.046 | 0.780 † 0.055 |
| Patient + drug + shuffled transition | 0.665 0.030 | 0.770 0.038 |
| Patient + drug + random transition | 0.676 0.046 | 0.773 0.046 |
| Patient + drug + predicted post-state | 0.685 0.048 | 0.781 0.054 |
| PerturbRx (Patient + drug + transition) | 0.692 0.055 | 0.787 0.055 |
Table 4: Ablation results on the TCGA-508.
| Method | Overall | Alpelisib ‡ | Buparlisib | Erlotinib ‡ | Fluorouracil ‡ | Gemcitabine ‡ | Paclitaxel ‡ | Ribociclib ‡ | Ruxolitinib | Tamoxifen | Trametinib ‡ |
| CODE-AE | 0.432 0.035 | 0.472 0.193 | 0.474 0.106 | 0.400 0.253 | 0.445 0.153 | 0.467 0.317 | 0.589 0.140 | 0.486 0.047 | 0.439 0.103 | 0.476 0.268 | 0.407 0.245 |
| DeepSADR | 0.637 0.024 | 0.587 0.079 | 0.595 0.074 | 0.300 0.217 | 0.304 0.196 | 0.462 0.349 | 0.511 0.065 | 0.497 0.082 | 0.762 0.158 | 0.719 0.206 | 0.430 0.258 |
| TransDRP | 0.470 0.026 | 0.418 0.124 | 0.438 0.107 | 0.367 † 0.183 | 0.508 † 0.373 | 0.463 0.359 | 0.540 0.232 | 0.525 † 0.047 | 0.342 0.133 | 0.181 0.262 | 0.640 0.123 |
| WISER | 0.492 † 0.031 | 0.527 † 0.047 | 0.503 † 0.145 | 0.400 0.384 | 0.616 0.138 | 0.465 † 0.214 | 0.554 † 0.174 | 0.576 0.084 | 0.575 † 0.209 | 0.595 † 0.337 | 0.537 † 0.322 |
| PerturbRx | 0.669 0.025 | 0.631 0.109 | 0.544 0.054 | 0.767 0.149 | 0.680 0.172 | 0.473 0.334 | 0.634 0.264 | 0.607 0.072 | 0.682 0.221 | 0.657 0.211 | 0.760 0.189 |
Table 5: Drug-stratified performance on the shared 10-drug PDX benchmark. Values are mean standard deviation across five model-disjoint outer folds. Best, second-best, and third-best results are indicated by bold , underline , and , respectively. Drug-specific AUPRC should be interpreted relative to the corresponding positive-class prevalence. ‡ indicates exact Tahoe-100M support by normalized drug name or canonicalized molecular structure.
Appendix figures & tables19 assets
Supplementary material from the paper’s appendix.
Appendix
| Symbol | Meaning |
| Source context defined by cell line and plate | |
| Intervention defined by drug and dose | |
| DMSO control population in context | |
| Treated population in context under intervention | |
| Frozen-encoder embeddings of and | |
| Learned intervention embedding for |
Table 6: Notation used in the PerturbRx formulation.
| Statistic | Value |
| Post-QC cells | 100.6M |
| Plates | 14 |
| Cell lines | 50 |
| Treated drugs | 379 |
| Dose levels | 0.05, 0.5, 5.0 M |
| Observed treated conditions | 65,218 |
Table 7: Statistics of the Tahoe-100M source-stage perturbation dataset. Observed treated conditions are defined as observed cell line, plate, drug, and dose combinations. Matched DMSO groups are defined by cell line and plate.
| Drug | Samples | Responders | Non-responders |
| Cisplatin | 40 | 20 | 20 |
| Fluorouracil | 28 | 12 | 16 |
| Gemcitabine | 46 | 23 | 23 |
| Sorafenib | 26 | 13 | 13 |
| Temozolomide | 46 | 23 | 23 |
| Total | 186 | 91 | 95 |
Table 8: Drug-level composition of TCGA-186. Response labels are relapse-time proxies derived from the processed WISER-style benchmark.
| Cancer type | Cis. | Flu. | Gem. | Sor. | Tem. | Total |
| Adrenocortical | 0 | 0 | 0 | 1 | 0 | 1 |
| Bladder | 0 | 0 | 4 | 0 | 0 | 4 |
| Brain | 0 | 0 | 0 | 0 | 42 | 42 |
| Cervical | 5 | 0 | 0 | 0 | 0 | 5 |
| Colorectal | 0 | 8 | 0 | 0 | 0 | 8 |
| Esophagus/Stomach | 1 | 20 | 0 | 0 | 0 | 21 |
Table 9: Cancer-type and drug composition of TCGA-186. Entries denote the number of unique patients associated with each cancer type and treatment. Cancer types were assigned using TCGA project metadata from the Genomic Data Commons.
| Drug | Episodes | ORR+ | ORR- | Prevalence | Tahoe match |
| Cisplatin | 151 | 120 | 31 | 0.795 | No |
| Docetaxel | 52 | 29 | 23 | 0.558 | Yes |
| Fluorouracil | 106 | 69 | 37 | 0.651 | Yes |
| Gemcitabine | 112 | 42 | 70 | 0.375 | Yes |
| Paclitaxel | 87 | 65 | 22 | 0.747 | Yes |
| Total | 508 | 325 | 183 | 0.640 | - |
Table 10: Drug-level composition of the TCGA-508 analytic cohort after expression matching. ORR-positive denotes complete or partial response, whereas ORR-negative denotes stable or progressive disease. An exact Tahoe match indicates agreement under harmonized drug-name and canonical-SMILES matching.
| Tumor type | Cisplatin | Docetaxel | Fluorouracil | Gemcitabine | Paclitaxel | Total |
| BRCA | 0 | 25 | 0 | 4 | 48 | 77 |
| PAAD | 1 | 0 | 10 | 66 | 0 | 77 |
| CESC | 62 | 0 | 4 | 1 | 2 | 69 |
| STAD | 5 | 2 | 51 | 0 | 1 | 59 |
| HNSC | 44 | 3 | 0 | 1 | 2 | 50 |
| BLCA | 10 | 3 | 0 | 17 | 1 | 31 |
Table 11: Tumor-type composition of the TCGA-508 benchmark across the five evaluated drugs. Entries denote treatment-episode counts for each tumor type-drug combination.
| Tumor type | Models | Episodes | Drugs | Response prevalence |
| BRCA | 38 | 227 | 6 | 0.401 |
| CM (SKCM) | 32 | 64 | 2 | 0.375 |
| CRC (COAD/READ) | 43 | 168 | 7 | 0.470 |
| NSCLC (LUAD/LUSC) | 25 | 124 | 5 | 0.387 |
| PDAC (PAAD) | 39 | 208 | 6 | 0.486 |
| Total | 177 | 791 | 10 | 0.434 |
Table 12: Tumor-type composition of the shared 10-drug PDX benchmark. Parenthetical labels denote the corresponding TCGA project codes. Response prevalence denotes the proportion of episodes labeled CR, PR, or SD.
| Tumor type | Alpelisib | Buparlisib | Erlotinib | Fluorouracil | Gemcitabine | Paclitaxel | Ribociclib | Ruxolitinib | Tamoxifen | Trametinib | Total |
| BRCA | 38 | 38 | 0 | 0 | 0 | 38 | 38 | 37 | 38 | 0 | 227 |
| PDAC | 36 | 35 | 0 | 0 | 34 | 0 | 35 | 33 | 0 | 35 | 208 |
| CRC | 42 | 40 | 0 | 41 | 1 | 0 | 42 | 1 | 0 | 1 | 168 |
| NSCLC | 25 | 25 | 25 | 0 | 0 | 24 | 25 | 0 | 0 | 0 | 124 |
| CM | 0 | 32 | 0 | 0 | 0 | 0 | 32 | 0 | 0 | 0 | 64 |
| Total | 141 | 170 | 25 | 41 | 35 | 62 | 172 | 71 | 38 | 36 | 791 |
Table 13: Tumor-drug composition of the shared 10-drug PDX benchmark. Entries denote treatment-episode counts for each tumor type-drug combination.
| Drug | Primary target / mechanism | Exact Tahoe match | Nearest Tahoe compound | ChemBERTa cosine |
| Alpelisib | PI3K | Yes | Sonidegib | 0.906 |
| Buparlisib | Pan-PI3K | No | Bimiralisib | 0.991 |
| Cisplatin | DNA cross-linking | No | Sodium Salicylate | 0.753 |
| Docetaxel | Microtubule stabilization | Yes | Docetaxel (Trihydrate) | 0.996 |
| Erlotinib | EGFR | Yes | XRK3F2 | 0.823 |
| Fluorouracil | TYMS / DNA–RNA incorporation | Yes | Methylthiouracil | 0.882 |
Table 14: Chemical coverage of target drugs in Tahoe-100M. Exact support indicates whether the target drug is present in Tahoe-100M by normalized drug name or canonicalized molecular structure. For each target drug, the nearest non-identical Tahoe compound is reported under ChemBERTa embedding cosine similarity and Morgan ECFP4 Tanimoto similarity. Exact target matches were excluded from nearest-neighbor searches. The two similarity measures are reported independently.
| Component | Configuration |
| Source-stage transition learning | |
| Cell encoder | Frozen scFoundation, |
| Drug encoder | Frozen ChemBERTa, |
| Intervention encoder | Input ; hidden/output dim |
| Transition predictor | Input , hidden dim , output dim |
| Dropout | |
Table 15: Architecture and hyperparameter details for PerturbRx.
Figure 4: Source-target embedding distribution diagnostics. A-C. Balanced joint PCA of TCGA bulk-tumor embeddings, Tahoe DMSO single-cell embeddings, and Tahoe cell-line centroids using A raw embeddings, B sample-wise -normalized embeddings, and C feature-wise standardized embeddings. D. Distribution of raw-embedding PC1 scores for TCGA patients and Tahoe DMSO cells. PCA was fit using equal numbers of TCGA patients and Tahoe cells.
| AUROC | AUPRC | |||||
| Drug | 0.05 | 0.5 | 5.0 | 0.05 | 0.5 | 5.0 |
| Cisplatin | 0.684 0.056 | 0.641 0.086 | 0.647 0.021 | 0.733 0.093 | 0.705 0.111 | 0.711 0.046 |
| Fluorouracil | 0.674 0.104 | 0.560 0.214 | 0.600 0.197 | 0.645 0.134 | 0.526 0.173 | 0.576 0.197 |
| Gemcitabine | 0.551 0.082 | 0.507 0.100 | 0.491 0.124 | 0.530 0.055 | 0.511 0.081 | 0.522 0.106 |
| Sorafenib | 0.753 0.213 | 0.667 0.162 | 0.740 0.201 | 0.801 0.174 | 0.704 0.160 | 0.773 0.172 |
| Temozolomide | 0.653 0.142 | 0.638 0.133 | 0.647 0.164 | 0.680 0.107 | 0.655 0.086 | 0.651 0.115 |
Table 16: Drug-wise dose sensitivity of PerturbRx on TCGA-186 under the fixed final response-model configuration. Values are mean standard deviation across five held-out-patient evaluations. Bold indicates the highest mean and underline the second-highest mean within each drug and metric. The inference dose is treated as a representation setting rather than a calibrated clinical exposure.
Figure 5: Transition-magnitude effect versus response-group size. Each point represents one cohort–drug stratum and shows the standardized difference in predicted transition magnitude between responders and non-responders. Negative values indicate smaller predicted transitions in responders. Effect estimates were more variable in sparsely represented strata, whereas larger strata tended to show negative or near-zero effects. This analysis is descriptive and is not used as primary evidence for a universal magnitude–response association.
| Drug representation | Distribution loss | Held-out condition cosine | |||
| ChemBERTa-zinc-base-v1 | MMD | 0.1 | 3.0 | 0.2099 | |
| ChemBERTa-zinc-base-v1 | Sinkhorn | 0.3 | 1.0 | 0 | 0.1906 |
| ChemBERTa-77M-MTR | MMD | 1.0 | 1.0 | 0.2074 | |
| ChemBERTa-77M-MTR | Sinkhorn | 0.3 | 3.0 | 0.1989 | |
| MoLFormer-XL | MMD | 1.0 | 1.0 | 0.1856 | |
| MoLFormer-XL | Sinkhorn | 1.0 | 3.0 | 0.1938 |
Table 17: Best source-stage configuration within each molecular representation and distributional-loss combination. Performance is measured by transition cosine similarity on randomly held-out source conditions; higher is better.
| AUROC | AUPRC | |||||
| Drug | 0.05 | 0.5 | 5.0 | 0.05 | 0.5 | 5.0 |
| Overall | 0.682 0.054 † | 0.692 0.055 | 0.684 0.062 | 0.783 0.063 † | 0.787 0.055 | 0.785 0.056 |
| Cisplatin | 0.686 0.166 | 0.631 0.168 † | 0.638 0.140 | 0.872 0.105 | 0.864 0.085 † | 0.869 0.075 |
| Docetaxel | 0.709 0.215 | 0.711 0.141 | 0.687 0.173 † | 0.816 0.171 | 0.802 0.112 | 0.772 0.150 † |
| Fluorouracil | 0.552 0.143 † | 0.562 0.116 | 0.568 0.076 | 0.714 0.120 † | 0.748 0.098 | 0.750 0.080 |
| Gemcitabine | 0.455 0.095 † | 0.472 0.095 | 0.460 0.108 | 0.387 0.053 | 0.395 0.052 | 0.385 0.058 † |
Table 18: Drug-wise dose sensitivity of PerturbRx on TCGA-508 under the fixed final response-model setting. Values are mean standard deviation across five patient-grouped evaluation folds. Bold indicates the best mean within each drug and metric across the three inference doses, underline indicates the second-best mean, and indicates the third-best mean. The inference dose is treated as a representation setting and is not calibrated to clinical treatment exposure.
| AUROC | AUPRC | |||||
| Drug | 0.05 | 0.5 | 5.0 | 0.05 | 0.5 | 5.0 |
| Fluorouracil | 0.680 0.172 † | 0.705 0.159 | 0.813 0.153 | 0.749 0.128 † | 0.782 0.120 | 0.831 0.142 |
| Buparlisib | 0.544 0.054 | 0.553 0.044 | 0.537 0.055 † | 0.559 0.056 | 0.538 0.061 | 0.529 0.051 † |
| Alpelisib | 0.631 0.109 † | 0.658 0.064 | 0.678 0.082 | 0.590 0.089 † | 0.591 0.107 | 0.678 0.066 |
| Ruxolitinib | 0.682 0.221 | 0.564 0.144 † | 0.738 0.182 | 0.361 0.268 | 0.312 0.245 † | 0.351 0.190 |
| Ribociclib | 0.607 0.072 | 0.559 0.074 † | 0.572 0.069 | 0.535 0.069 | 0.501 0.097 † | 0.508 0.069 |
Table 19: Drug-wise dose sensitivity of PerturbRx on the shared 10-drug PDX benchmark. Values are mean standard deviation across five model-disjoint outer folds. Bold indicates the best mean within each drug and metric across the three inference doses, underline indicates the second-best mean, and indicates the third-best mean. The tested doses are representation settings inherited from Tahoe-100M and should not be interpreted as calibrated PDX exposures.
| Benchmark | Drug | Patient+Drug | PerturbRx | AUROC |
| TCGA-508 | Cisplatin | 0.648 | 0.631 | -0.017 |
| Docetaxel | 0.636 | 0.711 | +0.075 | |
| Fluorouracil | 0.553 | 0.562 | +0.009 | |
| Gemcitabine | 0.462 | 0.472 | +0.010 | |
| Paclitaxel | 0.605 | 0.715 | +0.110 | |
| TCGA-186 | Cisplatin | 0.669 | 0.684 | +0.016 |
Table 20: Drug-specific incremental utility of the transferred transition representation across target benchmarks. Values report mean AUROC across the corresponding outer folds or evaluation seeds. AUROC is defined as . Positive values indicate improved discrimination after adding the predicted transition.
| Benchmark | Factor | ||
| TCGA-508 | Morgan ECFP4 Tanimoto | +0.900 | 0.083 |
| ChemBERTa cosine | +0.900 | 0.083 | |
| TCGA-186 | Morgan ECFP4 Tanimoto | +0.100 | 0.950 |
| ChemBERTa cosine | +0.500 | 0.450 | |
| PDX | Morgan ECFP4 Tanimoto | -0.067 | 0.865 |
| ChemBERTa cosine | -0.030 | 0.946 |
Table 21: Rank associations between drug-specific transition utility, measured by AUROC, and candidate explanatory factors. For chemical similarity, -values are based on two-sided exact permutation tests. PDX target-domain support variables use the same exact permutation framework over the ten-drug PDX benchmark.
Figure 6: PDX responder prevalence and drug-specific transition utility. Each point represents one drug in the shared 10-drug PDX benchmark. Transition utility is measured as the change in drug-specific AUROC after augmenting the static patient–drug representation with the predicted transition. The observed responder fraction was positively rank-associated with transition utility ( , two-sided exact permutation ). The two largest decreases occurred for Tamoxifen and Ruxolitinib, which also had the lowest observed responder fractions.