Conditional out-of-distribution generation for unpaired data using transfer VAE
Open accessThe DOI matches its Crossref record; 11 of the 14 references it cites were found in a registry or an index.
Abstract
Motivation: While generative models have shown great success in sampling high-dimensional samples conditional on low-dimensional descriptors (stroke thickness in MNIST, hair color in CelebA, speaker identity in WaveNet), their generation out-of-distribution poses fundamental problems due to the difficulty of learning compact joint distribution across conditions. The canonical example of the conditional variational autoencoder (CVAE), for instance, does not explicitly relate conditions during training and, hence, has no explicit incentive of learning such a compact representation. Results: We overcome the limitation of the CVAE by matching distributions across conditions using maximum mean discrepancy in the decoder layer that follows the bottleneck. This introduces a strong regularization both for reconstructing samples within the same condition and for transforming samples across conditions, resulting in much improved generalization. As this amount to solving a style-transfer problem, we refer to the model as transfer VAE (trVAE). Benchmarking trVAE on high-dimensional image and single-cell RNA-seq, we demonstrate higher robustness and higher accuracy than existing approaches. We also show qualitatively improved predictions by tackling previously problematic minority classes and multiple conditions in the context of cellular perturbation response to treatment and disease based on high-dimensional single-cell gene expression data. For generic tasks, we improve Pearson correlations of high-dimensional estimated means and variances with their ground truths from 0.89 to 0.97 and 0.75 to 0.87, respectively. We further demonstrate that trVAE learns cell-type-specific responses after perturbation and improves the prediction of most cell-type-specific genes by 65%. Availability and implementation: The trVAE implementation is available via github.com/theislab/trvae. The results of this article can be reproduced via github.com/theislab/trvae_reproducibility.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“Multiple distance measures for scRNA-seq have been explored by the single-cell community in recent years, including Wasserstein distance computed in an optimal transport framework22,23, maximum mean discrepancy24, neighborhood-based measures17,18, and energy distance (E-distance)20.”
From scPerturb: harmonized single-cell perturbation data (2024) · cites this paper as [24]
In the citing paper
“Contrary to scGEN, trVAE can address the issue of performing OOD predictions for multiple perturbations within a given dataset.”
From A mini-review on perturbation modelling across single-cell omic modalities (2024) · cites this paper as [53]
In the citing paper
“Aside from CPA, existing methods (Lotfollahi et al, 2020; Russkikh et al, 2020) such as scGen (Lotfollahi et al, 2019) have also been shown capable of predicting single‐cell perturbation responses when the dataset contains no combinatorial treatment or dose‐dependent perturbations.”
From Predicting cellular responses to complex perturbations in high‐throughput screens (2023) · cites this paper as Lotfollahi et al, 2020
Citations
Cites
Advances in neural information processing systems 7
(1996) · Computers & Mathematics with Applications · doi:10.1016/s0898-1221(96)90463-0 · Cited by 14,281
Advances in Neural Information Processing Systems 29
Teymur, O., Zygalakis, K., & Calderhead, B. (2016) · Cited by 13,887
(2012) · MPG.PuRe (Max Planck Society) · doi:10.5555/2188385.2188410 · Cited by 3,603
Deep generative modeling for single-cell transcriptomics
Lopez, R., Regier, J., Cole, M. B., Jordan, M. I., & Yosef, N. (2018) · Nature Methods · doi:10.1038/s41592-018-0229-2 · Cited by 2,530
A single-cell survey of the small intestinal epithelium
Haber, A. L., Biton, M., Rogel, N., Herbst, R. H., Shekhar, K., Smillie, C., Burgin, G., Delorey, T. M., Howitt, M. R., Katz, Y., et al. (2017) · Nature · doi:10.1038/nature24489 · Cited by 1,750
Multiplexed droplet single-cell RNA-sequencing using natural genetic variation
Kang, H. M., Subramaniam, M., Targ, S., Nguyen, M., Maliskova, L., McCarthy, E., Wan, E., Wong, S., Byrnes, L., Lanata, C. M., et al. (2018) · Nature Biotechnology · doi:10.1038/nbt.4042 · Cited by 1,101
Single-cell RNA-seq denoising using a deep count autoencoder
Eraslan, G., Simon, L. M., Mircea, M., Mueller, N. S., & Theis, F. J. (2019) · Nature Communications · doi:10.1038/s41467-018-07931-2 · Cited by 1,084
Self-citationscGen predicts single-cell perturbation responses
Lotfollahi, M., Wolf, F. A., & Theis, F. J. (2019) · Nature Methods · doi:10.1038/s41592-019-0494-8 · Cited by 661
Self-citationMassively multiplex chemical transcriptomics at single-cell resolution
Srivatsan, S. R., McFaline-Figueroa, J. L., Ramani, V., Saunders, L., Cao, J., Packer, J., Pliner, H. A., Jackson, D. L., Daza, R. M., Christiansen, L., et al. (2020) · Science · doi:10.1126/science.aax6234 · Cited by 401
Exploring single-cell data with deep multitasking neural networks
Amodio, M., van Dijk, D., Srinivasan, K., Chen, W. S., Mohsen, H., Moon, K. R., Campbell, A., Zhao, Y., Wang, X., Venkataswamy, M., et al. (2019) · Nature Methods · doi:10.1038/s41592-019-0576-7 · Cited by 297
Research works printed but not linked: 6
- Arjovsky, 2017, Proceedings of the 34th International Conference on Machine Learning, Volume 70 of Proceedings of Machine Learning Research, 214
- Dziugaite, 2015, Proceedings of the Thirty-First Conference on Uncertainty in Artificial Intelligence, UAI’15, 258
- Johansson, 2016, International Conference on Machine Learning, 3020
- Lopez, 2018, Advances in Neural Information Processing Systems, 6114
- Louizos, 2017, Advances in Neural Information Processing Systems, 6446
- Sohn, 2015, Learning structured output representation using deep conditional generative models, Advances in Neural Information Processing Systems, Vol. 28, 3483
Cited by
Items 101–131 of 131
inClust: a general framework for clustering that integrates data from multiple sourcesPreprint
Wang, L., Nie, R., Zhang, Z., Gu, W., Wang, S., Wang, A., Zhang, J., & Cai, J. (2022) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2022.05.27.493706 · Cited by 1
Building a single cell transcriptome-based coordinate system for cell ID with SUREPreprint
Zeng, F., & Han, J. (2024) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2024.11.13.623403 · Cited by 2
Beker, O., Deursen, S. V., Tarnow, M., Amador, D., Cheong, J. C., Nima, J. P., Robinson, M. D., Woappi, Y., & Dumitrascu, B. (2024) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2024.12.23.630186 · Cited by 2
Powerful and accurate case-control analysis of spatial molecular dataPreprint
Reshef, Y., Sood, L., Curtis, M. L., Rumker, L., Stein, D. J., Palshikar, M. G., Nayar, S., Filer, A. D., Jonsson, A. H., Korsunsky, I., et al. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.02.07.637149 · Cited by 1
Zhu, H., Asiaee, A., Azinfar, L., Li, J., Liang, H., Irajizad, E., Do, K.-A., & Long, J. P. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.03.06.641935 · Cited by 1
“While R 2 has become the de facto standard for evaluating gene expression prediction models [18, 17, 19, 32, 29, 5], it presents several limitations in the context of cellular perturbation experiments.”
Radig, J., Droit, R., Doncevic, D., Li, A., Bui, D. T., Herfurth, L., Kühn, T., & Herrmann, C. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.06.23.661046 · Cited by 1
Roseth, I. A., Hatscher, L., Schiller, C., Johnstuen, H. S. S., Slørdahl, T., Hov, H., Schapiro, D., & Standal, T. (2025) · bioRxiv · doi:10.1101/2025.09.08.674473 · Cited by 1
Unbalanced Perturbation Dynamics For Cell Fate DesignPreprint
Peng, Q., Wang, Y., Li, J., Wang, X., Xiao, Y., & Zhou, P. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.06.30.735555 · Cited by 1
Transfer Learning of Clinical Outcomes with Molecular Data, Principles and PerspectivesPreprint
Kowald, A., Barrantes, I., Möller, S., Palmer, D. H., Escobar, H. M., & Fuellen, G. (2021) · Preprints.org · doi:10.20944/preprints202110.0207.v1
“Conditional out-of-distribution transfer learning by Lotfollahi et al. (2020) employs a transfer variational autoencoder (trVAE), which enables the transfer of conditions across domains.”
BackgroundInfluentialSimple Causal Relationships in Gene Expression Discovered through Deep Learned Collective VariablesPreprint
Wang, C.-H., Vetsigian, K., Lin, C., Firth, F., Bradley, G., Granovsky, L., & England, J. L. (2023) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2023.01.18.524617
“The use of autoencoders [9], UMAP [10, 11], and other methods of dimensionality reduction [12, 13] have demonstrated that it is possible to discover relationships in statistical covariation of many genes that can be used, for example, to distinguish different cell types and lineages [14], to predict mean gene expressi…”
MethodITSMatch: Improved Safe Semi-Supervised Text Classification Under Class Distribution Mismatch
Zeng, K., Li, J., Xu, Y., Zhang, X., Ge, L., Zeng, N., & Qu, Y. (2023) · IEEE Transactions on Consumer Electronics · doi:10.1109/tce.2023.3323982 · Cited by 1
“Similarly, Lotfollahi et al. assess on OOD data by calculating the reconstruction loss of unlabeled text [49].”
Molecular Generator for Multi-objective Optimization Based on the Pareto Algorithm
Feng, D. (2023) · MEDS Clinical Medicine · doi:10.23977/medsc.2023.040809
“[21] The model can generate similar but not identical compounds by adding random noise meanwhile considering the physicochemical properties and structure of the training compounds.”
BackgroundContinual integration of single-cell multimodal data with MIRACLE
Zhou, J., Wang, J., Hu, S., Kan, T., Feng, C., Qiang, X., Dong, G., Shi, J., Liu, R., Bo, X., et al. (2026) · Nature Computational Science · doi:10.1038/s43588-026-01030-9 · Cited by 1
“SCALEX-projection trained on the first batch and projected the remaining batches (embedding dimension 32). trVAE . trVAE 74,75 supports a single modality, so we concatenated RNA and ADT. trVAE+scArches added scArches transfer learning, training a reference on the first batch and adaptors on the…”
Causal models and prediction in cell line perturbation experiments
Long, J. P., Yang, Y., Shimizu, S., Pham, T., & Do, K.-A. (2025) · BMC Bioinformatics · doi:10.1186/s12859-024-06027-7
Hu, H., Guo, Y., Ge, F., Yin, H. R., Zhang, H., Zhou, Z., Yan, F., Ye, Q., Wu, J., Cao, J., et al. (2025) · Advanced Science · doi:10.1002/advs.202410790
Advances in single-cell perturbation modeling
Qi, X., Zhao, L., Chen, R., & Zhao, Y. C. (2025) · Chinese Science Bulletin (Chinese Version) · doi:10.1360/csb-2025-0682
Cheng, Y., & Zhang, X. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.09.19.677191
Hausmann, F., Ergen, C., Khatri, R., Marouf, M., Hänzelmann, S., Gagliani, N., Huber, S., Machart, P., & Bonn, S. (2023) · Figshare · doi:10.6084/m9.figshare.24172161
Hausmann, F., Ergen, C., Khatri, R., Marouf, M., Hänzelmann, S., Gagliani, N., Huber, S., Machart, P., & Bonn, S. (2023) · Figshare · doi:10.6084/m9.figshare.24172167
Hausmann, F., Ergen, C., Khatri, R., Marouf, M., Hänzelmann, S., Gagliani, N., Huber, S., Machart, P., & Bonn, S. (2023) · Figshare · doi:10.6084/m9.figshare.24172164
Joint modeling of cellular heterogeneity and condition effects with scPCA in single-cell RNA-seq
Vöhringer, H. S., & Dietrich, S. (2026) · Communications Biology · doi:10.1038/s42003-026-09651-6
“The performance of scPCA is evaluated using both simulated and real scRNA-seq datasets, specifically in terms of its ability to model batch effects, extract cell-type-specific factors, and handle multiple time-point data, among other aspects.”
Unsupervised representation learning for single-cell transcriptomic and epigenomic data
Raimundo, F. (2022) · theses.fr (ABES) · doi:10.70675/760cd5d0z17c6z4f4ezb476z68bd387daf22
Dhaka, B., Gao, J., & Ideker, T. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.04.23.720505 · Cited by 1
“One set of methods uses generative modelling based on variational autoencoders (VAEs), which represent perturbed cell states as a function of baseline expression, perturbation inputs, and optional contextual covariates (e.g., cell type) 13 .”
Elucidating the Design Space of Generative Models for Single-Cell Perturbation PredictionPreprint
Bhattacharya, S., Gensbigler, C., Karim, S., & Lees, J. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.06.15.732063
“…the canonical Gaussian-latent, Negative Binomial-likelihood VAE recipe, elaborated by the conditional-VAE thread of scGen, trVAE, and CPA (20; 21; 22) and the latent-diffusion thread of scDiffusion (24), CFGen (30), and scLDM (29) with flow-matching priors (16; 31); the Hurdle head we…”
Du, Z., Zhang, X., Zhang, Q. P., Li, J., Cao, Z., Gao, G., Lin, T., Wang, D., & Gao, S. (2026) · Genome Research · doi:10.1101/gr.281629.125
Unify learns cellular evolution with universal multimodal embeddings
Zhong, H., Han, W., Cui, G., Gómez-Cabrero, D., Tegner, J. N., Gao, X., & Aranda, M. (2026) · Nature Communications · doi:10.1038/s41467-026-76230-y · Cited by 2
“Additionally, existing integration approaches typically produce either corrected expression pro fi les restricted to one-to-one orthologous genes 14 – 17 or integrated embeddings 6,10,18 – 20 that lack direct interpretability.”
Wu, M., Lu, J., & Ge, Q. (2026) · Cancer Genetics · doi:10.1016/j.cancergen.2026.08.002
Reliable single-cell perturbations explain and improve model performancePreprint
Wang, X., Kuipers, J., Hugi, F., Platt, R. J., & Beerenwinkel, N. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.08.11.744177 · Cited by 5
Yao, S., & Chen, X. (2026) · Frontiers in Pharmacology · doi:10.3389/fphar.2026.1962508
“TrVAE subsequently extended this framework to unpaired biological datasets, enabling conditional prediction when matched control and treated samples are unavailable (Lotfollahi et al., 2020).”
Accurate and well-powered case–control analysis of spatial molecular data
Reshef, Y., Sood, L., Curtis, M. L., Rumker, L., Stein, D. J., Palshikar, M. G., Nayar, S., Filer, A. D., Jonsson, A. H., Korsunsky, I., et al. (2026) · Nature Methods · doi:10.1038/s41592-026-03236-1 · Cited by 2
“…we did not https://doi.org/10.1038/s41592-026-03236-1 have access to segmented cells; and in the UC dataset, we were unable to apply CellCharter because its preprocessing step for CODEX data, which integrates expression profiles of all cells using trVAE 62 , exceeded our runtime limit of 3 days.”
InfluentialDART: Dual-Path Adaptive Residual Transformer for Virtual Single-cell Drug Perturbation Prediction
Jiang, C., Zheng, R., Liang, S., & Tao, S. (2026) · International Conference of Data and Information Systems (DIS) · doi:10.1109/dis72268.2026.11725700
“Recent advances in single-cell RNA sequencing (scRNA-seq) have enabled virtual cell models that map initial cellular states and drug information to post-perturbation expression profiles [1], [2].”
Metadata
- Authors
- Lotfollahi, M., Naghipourfar, M., Theis, F. J., & Wolf, F. A.
- Affiliations
- Helmholtz Munich (DE)
- Corresponding author
- Theis, Fabian J
- Wolf, F Alexander
- Venue
- Bioinformatics
- Publisher
- Oxford University Press (OUP)
- Published
- 2020-12
- Volume
- 36
- Issue
- Supplement_2
- Pages
- i610-i617
- ISSN
- 1367-4803 (ISSN-L)
- 1367-4811
- OpenAlex
- W3117464799
- Cited by
- 130
- References
- 14
- Citations by year
- 2026: 25 · 2025: 31 · 2024: 24 · 2023: 25 · 2022: 12 · 2021: 12 · 2020: 1
- FWCI
- 4.05
- Type
- journal article
- Language
- en
- Open access
- Licence
- Locations
- Bioinformatics (publishedVersion)
- PubMed (publishedVersion)
- Site cant be reached (submittedVersion, other-oa)
- PuSH - Publication Server of Helmholtz Zentrum München (submittedVersion)
- Funders
- BMBF (01IS18036A, 01IS18053A)
- German Research Foundation (ZT-I-0007)
- Chan Zuckerberg Initiative DAF (182835)
- Joachim Herz Stiftung
- Topics
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Single-cell and spatial transcriptomics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Generative Adversarial Networks and Image Synthesis
- Physical Sciences › Computer Science › Artificial Intelligence › Gaussian Processes and Bayesian Inference
- Keywords
- conditional variational autoencoder, maximum mean discrepancy, style transfer, unpaired data, single-cell RNA sequencing, cell type-specific gene expression, perturbation response prediction, minority class, generative models
- MeSH terms
- Exome Sequencing, RNA-Seq, Reproducibility of Results
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.