pith. sign in

arxiv: 2605.31522 · v1 · pith:KAFJW2O6new · submitted 2026-05-29 · 💻 cs.LG · q-bio.GN· q-bio.QM

Chem-PerturBridge: a harmonized compendium of small molecule perturbation transcriptomic effects

Artur Sza{\l}ata , Olga Novitskaia , Maiia Shulman , Matthew Mella , Altynbek Zhubanchaliyev , Fabian J. Theis This is my paper

Pith reviewed 2026-06-28 23:29 UTC · model grok-4.3

classification 💻 cs.LG q-bio.GNq-bio.QM
keywords perturbation transcriptomicsdata harmonizationcompound representation learningsmall molecule effectsL1000cross-dataset agreementpretraining embeddingsOP3 evaluation
0
0 comments X

The pith

Harmonized transcriptomic perturbation data across eight assays improves compound embedding performance over single-source baselines.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper assembles Chem-PerturBridge by aligning more than 1.25 million transcriptomic samples from small-molecule perturbations across eight assay types, 136 contexts, and 37,000 compounds. It reports that fine-grained log-fold-change rankings agree weakly between matched conditions from different datasets, yet the direction of change remains stable and exceeds random baselines. Pretraining compound embeddings on the full resource raises performance in compound-held-out OP3 splits relative to L1000-only embeddings, Morgan fingerprints, and a descriptor-free baseline. Models trained on the resource also match or exceed non-pretrained models when tested across eleven external molecule-holdout datasets.

Core claim

Chem-PerturBridge supplies a harmonized collection of over 37k compounds and 1.25M transcriptomic samples with standardized identifiers and replicate-aware condition-level effects. Embeddings pretrained on this collection improve over L1000-only embeddings and Morgan fingerprints under a compound-held-out OP3 evaluation split, and models trained on the resource outperform or match those trained without it in molecule-holdout tests on eleven datasets.

What carries the argument

Chem-PerturBridge, the harmonized multi-dataset resource that standardizes identifiers, metadata, and replicate-aware condition-level effects across eight assay types.

If this is right

  • Embeddings pretrained on Chem-PerturBridge improve over L1000-only embeddings, Morgan fingerprints, and the descriptor-free baseline across metrics in compound-held-out OP3 splits.
  • Models trained on Chem-PerturBridge outperform or match models that are not in molecule-holdout evaluations across eleven datasets.
  • Matched same-compound conditions exhibit weak agreement in fine-grained logFC rankings and magnitudes across most dataset pairs.
  • LogFC direction agreement across dataset pairs is substantially more stable and usually exceeds same-context different-compound baselines.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Direction agreement alone may support tasks that care only about up- or down-regulation rather than exact effect size.
  • Extending the same harmonization approach to additional omics layers could produce still richer compound representations.
  • Similar standardization efforts applied to other fragmented biological assay collections might yield comparable pretraining gains.

Load-bearing premise

Harmonization across technologies, metadata conventions, controls, doses, and preprocessing pipelines produces biologically comparable condition-level effects that preserve true perturbation signals rather than artifacts.

What would settle it

A compound-held-out evaluation in which embeddings pretrained on Chem-PerturBridge show no improvement over L1000-only embeddings or Morgan fingerprints on the OP3 split would falsify the utility claim.

Figures

Figures reproduced from arXiv: 2605.31522 by Altynbek Zhubanchaliyev, Artur Sza{\l}ata, Fabian J. Theis, Maiia Shulman, Matthew Mella, Olga Novitskaia.

Figure 1
Figure 1. Figure 1: Cross-dataset overlap and logFC agreement across Chem-PerturBridge. DEG￾restricted logFC correlation is generally weak, often at or below a within-dataset same-context different-compound baseline. Cross-dataset logFC direction agreement is substantially stronger and usually outperforms the baseline. Cross-dataset retrieval is generally better than random, but within-dataset same-context baseline similarity… view at source ↗
Figure 2
Figure 2. Figure 2: Within-dataset replicate agreement. Each row corresponds to a dataset from [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Three-stage harmonization and evaluation pipeline in Chem-PerturBridge. Stage 1 (Standardize & Harmonize): heterogeneous source assays enter through three streams: raw single￾cell data (Stream A: Tahoe-100M, sci-Plex), sample or pseudobulk counts (Stream B: OP3, CIGS, Novartis, DRUG-seq, GDPx2, VCPI, DILImap), and normalized targeted profiles (Stream C: L1000 Phase I/II) each routed through assay-specific … view at source ↗
Figure 4
Figure 4. Figure 4: Perturbation strength is associated with cross-dataset signature similarity. Each point is a matched condition-dose unit aggregated from matched sample-pair similarity scores, defined by dataset pair, cell context, time, PubChem CID, matched dose pair, and matched condition key. The x-axis shows log10 mean absolute moderated t across the two matched datasets, used as a proxy for perturbation strength. The … view at source ↗
Figure 5
Figure 5. Figure 5: Distribution of within-dataset replicate agreement for the DEG-restricted metrics in [PITH_FULL_IMAGE:figures/full_fig_p027_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Distribution of within-dataset all-gene replicate logFC Spearman agreement. For each dataset, the distribution of Spearman correlations between replicate-level logFC signatures is shown alongside the corresponding same-context different-compound baseline distribution. This analysis uses all shared genes in the replicate comparison rather than restricting to differentially expressed genes. The plot compleme… view at source ↗
read the original abstract

Large perturbation models require training data encompassing chemical, cellular, and assay diversity. Current transcriptomic resources for small-molecule modeling, however, are fragmented across technologies, metadata conventions, controls, doses, and preprocessing pipelines. We introduce Chem-PerturBridge, a harmonized multi-dataset resource comprising over 37k compounds, 136 cellular contexts, and 1.25M transcriptomic samples across eight assay types, with standardized identifiers, metadata, and replicate-aware condition-level effects. We use the resource to evaluate matched-condition agreement across datasets and replicate agreement within datasets. Matched same-compound conditions generally show weak agreement in fine-grained logFC rankings and magnitudes across most dataset pairs, often falling below same-context different-compound baselines. In contrast, logFC direction agreement is substantially more stable and usually exceeds these baselines. We further evaluate Chem-PerturBridge as a pretraining resource for compound representation learning. Under a compound-held-out OP3 evaluation split, embeddings pretrained on Chem-PerturBridge improve over L1000-only embeddings, Morgan fingerprints, and the descriptor-free OP3 baseline across metrics. An extensive molecule-holdout evaluation across 11 datasets further shows that models trained on Chem-PerturBridge outperform or match those that are not. Chem-PerturBridge therefore supports both diagnostic evaluation of cross-dataset signature agreement and model-oriented reuse of heterogeneous perturbation transcriptomic data.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces Chem-PerturBridge, a harmonized compendium aggregating over 1.25M transcriptomic samples from small-molecule perturbations across 37k compounds, 136 cellular contexts, and eight assay types, with standardized identifiers, metadata, and replicate-aware condition-level effects. It reports that matched same-compound conditions exhibit weak agreement in fine-grained logFC rankings and magnitudes across most dataset pairs (often below same-context different-compound baselines), while logFC direction agreement is more stable and exceeds baselines. The resource is evaluated as a pretraining corpus, with compound-held-out OP3 splits showing that pretrained embeddings outperform L1000-only embeddings, Morgan fingerprints, and descriptor-free baselines; an additional molecule-holdout evaluation across 11 datasets indicates that models trained on the harmonized data outperform or match non-pretrained counterparts.

Significance. If the harmonization produces biologically comparable signals, the resource would constitute a substantial advance for training large perturbation models, enabling better compound representation learning across heterogeneous transcriptomic assays. The agreement diagnostics provide a concrete benchmark for cross-technology comparability, and the pretraining results demonstrate empirical utility for downstream tasks in drug discovery and systems biology.

major comments (2)
  1. [Abstract] Abstract: the central pretraining claim (improved embeddings under compound-held-out OP3 and across 11 molecule-holdout datasets) rests on the assumption that harmonized condition-level effects preserve transferable biological signals rather than technology-specific artifacts. This assumption is directly challenged by the reported weak logFC ranking and magnitude agreement (frequently below different-compound baselines), which the manuscript does not resolve with additional controls or sensitivity analyses showing that pretraining gains survive removal of non-comparable features.
  2. [Abstract] Abstract and evaluation sections: the manuscript reports concrete agreement metrics and pretraining gains but provides no methodological details on statistical tests for agreement, replicate-aware aggregation procedures, dose/metadata standardization steps, or ablation of harmonization choices. Without these, it is impossible to determine whether the observed weak agreement or the pretraining improvements are robust to alternative processing decisions.
minor comments (1)
  1. [Abstract] The abstract would benefit from explicit mention of the number of datasets contributing to the 1.25M samples and the precise definition of the OP3 evaluation split.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which highlight important considerations for the interpretation and reproducibility of our work. We respond to each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central pretraining claim (improved embeddings under compound-held-out OP3 and across 11 molecule-holdout datasets) rests on the assumption that harmonized condition-level effects preserve transferable biological signals rather than technology-specific artifacts. This assumption is directly challenged by the reported weak logFC ranking and magnitude agreement (frequently below different-compound baselines), which the manuscript does not resolve with additional controls or sensitivity analyses showing that pretraining gains survive removal of non-comparable features.

    Authors: We agree that the weak agreement in logFC rankings and magnitudes raises valid questions about the degree of biological comparability across datasets. At the same time, the molecule-holdout evaluation across 11 datasets provides direct empirical evidence that pretraining on the harmonized resource yields improved or matched performance relative to non-pretrained baselines, which would be unlikely if the signals were dominated by non-transferable artifacts. Direction agreement being consistently above baseline further supports preservation of some biologically relevant information. We will add a dedicated discussion paragraph addressing the relationship between agreement metrics and pretraining utility, and we will perform a limited sensitivity check (e.g., restricting pretraining to direction-only or high-agreement feature subsets) to test robustness of the gains. revision: partial

  2. Referee: [Abstract] Abstract and evaluation sections: the manuscript reports concrete agreement metrics and pretraining gains but provides no methodological details on statistical tests for agreement, replicate-aware aggregation procedures, dose/metadata standardization steps, or ablation of harmonization choices. Without these, it is impossible to determine whether the observed weak agreement or the pretraining improvements are robust to alternative processing decisions.

    Authors: We accept this criticism. The current manuscript is insufficiently explicit on these processing choices. In the revised version we will expand the Methods section with: (i) the exact statistical tests and multiple-testing corrections applied to agreement metrics, (ii) the replicate-aware aggregation algorithm and its parameters, (iii) the full sequence of dose, metadata, and identifier standardization steps, and (iv) any ablations or sensitivity checks performed on harmonization decisions. These additions will enable readers to evaluate robustness directly. revision: yes

Circularity Check

0 steps flagged

No circularity: resource construction and empirical evaluation only

full rationale

The paper describes construction of a harmonized transcriptomic resource (Chem-PerturBridge) and reports empirical results from pretraining embeddings and molecule-holdout evaluations across datasets. No derivation chain, equations, or first-principles predictions exist. All claims rest on standard data processing, held-out splits, and baseline comparisons that are independent of any internal fit or self-citation reduction. The weak cross-dataset logFC agreement noted in the abstract is an empirical observation, not a circular step. This is a self-contained empirical study with no load-bearing reductions to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central contribution rests on the domain assumption that heterogeneous transcriptomic assays can be meaningfully harmonized; no free parameters or new physical entities are introduced.

axioms (1)
  • domain assumption Transcriptomic perturbation data from disparate assays, doses, and preprocessing pipelines can be standardized into comparable condition-level effects
    Invoked to justify creation of the multi-dataset resource and subsequent agreement and pretraining evaluations.

pith-pipeline@v0.9.1-grok · 5820 in / 1247 out tokens · 27625 ms · 2026-06-28T23:29:41.016761+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

61 extracted references · 46 canonical work pages

  1. [1]

    Crawford, David Peck, Joshua W

    Justin Lamb, Emily D. Crawford, David Peck, Joshua W. Modell, Irene C. Blat, Matthew J. Wrobel, Jim Lerner, Jean-Philippe Brunet, Aravind Subramanian, Kenneth N. Ross, Michael Reich, Haley Hieronymus, Guo Wei, Scott A. Armstrong, Stephen J. Haggarty, Paul A. Clemons, Ru Wei, Steven A. Carr, Eric S. Lander, and Todd R. Golub. The Connectivity Map: Using Ge...

    work page doi:10.1126/science.1132939 1929

  2. [2]

    Corsello, David D

    Aravind Subramanian, Rajiv Narayan, Steven M. Corsello, David D. Peck, Ted E. Natoli, Xiaodong Lu, Joshua Gould, John F. Davis, Andrew A. Tubelli, Jacob K. Asiedu, David L. Lahr, Jodi E. Hirschman, Zihan Liu, Melanie Donahue, Bina Julian, Mariya Khan, David Wadden, Ian C. Smith, Daniel Lam, Arthur Liberzon, Courtney Toder, Mukta Bagul, Marek Orzechowski, ...

    work page doi:10.1016/j.cell 2017

  3. [3]

    Keenan, Sherry L

    Alexandra B. Keenan, Sherry L. Jenkins, Kathleen M. Jagodnik, Simon Koplev, Edward He, Denis Torre, Zichen Wang, Anders B. Dohlman, Moshe C. Silverstein, Alexander Lachmann, Maxim V . Kuleshov, Avi Ma’ayan, Vasileios Stathias, Raymond Terryn, Daniel Cooper, Michele Forlin, Amar Koleti, Dusica Vidovic, Caty Chung, Stephan C. Schürer, Jouzas Vasiliauskas, M...

    work page doi:10.1016/j.cels.2017.11.001 2018

  4. [4]

    Alexander Wolf, and Fabian J

    Mohammad Lotfollahi, F. Alexander Wolf, and Fabian J. Theis. scGen predicts single- cell perturbation responses.Nature Methods, 16(8):715–721, August 2019. ISSN 1548-

    2019

  5. [5]

    URL https://www.nature.com/articles/ s41592-019-0494-8

    doi: 10.1038/s41592-019-0494-8. URL https://www.nature.com/articles/ s41592-019-0494-8

    work page doi:10.1038/s41592-019-0494-8

  6. [6]

    Ibarra, Sanjay R

    Mohammad Lotfollahi, Anna Klimovskaia Susmelj, Carlo De Donno, Leon Hetzel, Yuge Ji, Ignacio L. Ibarra, Sanjay R. Srivatsan, Mohsen Naghipourfar, Riza M. Daza, Beth Martin, Jay Shendure, Jose L. McFaline-Figueroa, Pierre Boyeau, F. Alexander Wolf, Nafissa Yakubova, Stephan Günnemann, Cole Trapnell, David Lopez-Paz, and Fabian J. Theis. Predicting cellular...

    work page doi:10.15252/msb.202211517 2023

  7. [7]

    Schmon, Marcel Nassar, Bła ˙zej Osi ´nski, Ridvan Eksi, Zichao Yan, Rory Stark, Kun Zhang, and Thore Graepel

    Yan Wu, Esther Wershof, Sebastian M. Schmon, Marcel Nassar, Bła ˙zej Osi ´nski, Ridvan Eksi, Zichao Yan, Rory Stark, Kun Zhang, and Thore Graepel. PerturBench: Benchmarking Machine Learning Models for Cellular Perturbation Analysis. October 2025. URL https: //openreview.net/forum?id=PPPDuyiZaG

    2025

  8. [8]

    In silico bi- ological discovery with large perturbation models.Nature Computational Science, 5(11): 1029–1040, November 2025

    Djordje Miladinovic, Tobias Höppe, Mathieu Chevalley, Andreas Georgiou, Lachlan Stuart, Arash Mehrjou, Marcus Bantscheff, Bernhard Schölkopf, and Patrick Schwab. In silico bi- ological discovery with large perturbation models.Nature Computational Science, 5(11): 1029–1040, November 2025. ISSN 2662-8457. doi: 10.1038/s43588-025-00870-1. URL https://www.nat...

    work page doi:10.1038/s43588-025-00870-1 2025

  9. [9]

    Schmacke, Zihe Zheng, Xinyue Zhang, Simrah Khan, Ina Rothenaigner, Juliane Tschuck, Kamyar Hadian, Veit Hornung, and Fabian J

    Yuge Ji, Alejandro Tejada-Lapuerta, Niklas A. Schmacke, Zihe Zheng, Xinyue Zhang, Simrah Khan, Ina Rothenaigner, Juliane Tschuck, Kamyar Hadian, Veit Hornung, and Fabian J. Theis. Scalable and universal prediction of cellular phenotypes enables in silico experiments, Septem- ber 2025. URL https://www.biorxiv.org/content/10.1101/2024.08.12.607533v3. ISSN: ...

    work page doi:10.1101/2024.08.12.607533v3 2025

  10. [10]

    Williams, Kenneth McCue, Lorian Schaeffer, and Barbara Wold

    Ali Mortazavi, Brian A. Williams, Kenneth McCue, Lorian Schaeffer, and Barbara Wold. Mapping and quantifying mammalian transcriptomes by RNA-Seq.Nature Methods, 5(7):621– 628, July 2008. ISSN 1548-7105. doi: 10.1038/nmeth.1226. URL https://www.nature. com/articles/nmeth.1226

    work page doi:10.1038/nmeth.1226 2008

  11. [11]

    Crawford, Kenneth N

    David Peck, Emily D. Crawford, Kenneth N. Ross, Kimberly Stegmaier, Todd R. Golub, and Justin Lamb. A method for high-throughput gene expression signature analysis.Genome Biology, 7(7):R61, July 2006. ISSN 1474-760X. doi: 10.1186/gb-2006-7-7-r61. URL https://doi.org/10.1186/gb-2006-7-7-r61

    work page doi:10.1186/gb-2006-7-7-r61 2006

  12. [12]

    Chaoyang Ye, Daniel J. Ho, Marilisa Neri, Chian Yang, Tripti Kulkarni, Ranjit Rand- hawa, Martin Henault, Nadezda Mostacci, Pierre Farmer, Steffen Renner, Robert Ihry, Le- andra Mansur, Caroline Gubser Keller, Gregory McAllister, Marc Hild, Jeremy Jenkins, and Ajamete Kaykas. DRUG-seq for miniaturized high-throughput transcriptome profil- ing in drug disc...

    2018

  13. [13]

    URL https://www.nature.com/articles/ s41467-018-06500-x

    doi: 10.1038/s41467-018-06500-x. URL https://www.nature.com/articles/ s41467-018-06500-x. 11

    work page doi:10.1038/s41467-018-06500-x

  14. [14]

    Xu, Lijun Huang, Dale Guo, Guanbin Zhang, Zhi Xie, Yun Deng, Junquan Xu, and Dong Wang

    Lei Xiang, Yumei Wang, Wei Shao, Qingzhou Li, Xiankuo Yu, Mingming Wei, Yu Gui, Shengrong Li, Pan Qin, Chao Hu, Guochen Zhang, Xianwen Zhang, Jiawen Wang, Yingying Li, Jun An, Yan Luo, Yile Liao, Jinghong Deng, Xinran Tai, Richard Y . Xu, Lijun Huang, Dale Guo, Guanbin Zhang, Zhi Xie, Yun Deng, Junquan Xu, and Dong Wang. High-throughput profiling of chemi...

    work page doi:10.1038/s41592-025-02781-5 1954

  15. [15]

    Srivatsan, José L

    Sanjay R. Srivatsan, José L. McFaline-Figueroa, Vijay Ramani, Lauren Saunders, Junyue Cao, Jonathan Packer, Hannah A. Pliner, Dana L. Jackson, Riza M. Daza, Lena Christiansen, Fan Zhang, Frank Steemers, Jay Shendure, and Cole Trapnell. Massively multiplex chemical transcriptomics at single-cell resolution.Science, 367(6473):45–51, January 2020. doi: 10.11...

    work page doi:10.1126/science.aax6234 2020

  16. [16]

    Ritchie, Belinda Phipson, Di Wu, Yifang Hu, Charity W

    Matthew E. Ritchie, Belinda Phipson, Di Wu, Yifang Hu, Charity W. Law, Wei Shi, and Gordon K. Smyth. limma powers differential expression analyses for RNA-sequencing and microarray studies.Nucleic Acids Research, 43(7):e47–e47, April 2015. ISSN 0305-1048. doi: 10.1093/nar/gkv007. URLhttps://dx.doi.org/10.1093/nar/gkv007

    work page doi:10.1093/nar/gkv007 2015

  17. [17]

    Clark, Yan Ren, Shana White, Rashid Karim, Huan Xu, Jacek Biesiada, Mark F

    Marcin Pilarczyk, Mehdi Fazel-Najafabadi, Michal Kouril, Behrouz Shamsaei, Juozas Vasil- iauskas, Wen Niu, Naim Mahi, Lixia Zhang, Nicholas A. Clark, Yan Ren, Shana White, Rashid Karim, Huan Xu, Jacek Biesiada, Mark F. Bennett, Sarah E. Davidson, John F. Re- ichard, Kurt Roberts, Vasileios Stathias, Amar Koleti, Dusica Vidovic, Daniel J. B. Clarke, Stepha...

    work page doi:10.1038/s41467-022-32205-3 2022

  18. [18]

    Monteiro, Kathleen M

    Zichen Wang, Caroline D. Monteiro, Kathleen M. Jagodnik, Nicolas F. Fernandez, Gregory W. Gundersen, Andrew D. Rouillard, Sherry L. Jenkins, Axel S. Feldmann, Kevin S. Hu, Michael G. McDermott, Qiaonan Duan, Neil R. Clark, Matthew R. Jones, Yan Kou, Troy Goff, Holly Wood- land, Fabio M. R. Amaral, Gregory L. Szeto, Oliver Fuchs, Sophia M. Schüssler-Fioren...

    work page doi:10.1038/ncomms12846 2016

  19. [19]

    Wilhite, Pierre Ledoux, Carlos Evangelista, Irene F

    Tanya Barrett, Stephen E. Wilhite, Pierre Ledoux, Carlos Evangelista, Irene F. Kim, Maxim Tomashevsky, Kimberly A. Marshall, Katherine H. Phillippy, Patti M. Sherman, Michelle Holko, Andrey Yefanov, Hyeseung Lee, Naigong Zhang, Cynthia L. Robertson, Nadezhda Serova, Sean Davis, and Alexandra Soboleva. NCBI GEO: archive for functional genomics data sets—up...

    work page doi:10.1093/nar/gks1193 2013

  20. [20]

    Ubas, Richard de Borja, Valentine Svensson, Nicole Thomas, Neha Thakar, Ian Lai, Aidan Winters, Umair Khan, Matthew G

    Jesse Zhang, Airol A. Ubas, Richard de Borja, Valentine Svensson, Nicole Thomas, Neha Thakar, Ian Lai, Aidan Winters, Umair Khan, Matthew G. Jones, Vuong Tran, Joseph Pangallo, Efthymia Papalexi, Ajay Sapre, Hoai Nguyen, Oliver Sanderson, Maria Nigos, Olivia Kaplan, Sarah Schroeder, Bryan Hariadi, Simone Marrujo, Crina Curca Alec Salvino, Guillermo Gal- l...

    work page doi:10.1101/2025.02.20.639398v1 2025

  21. [21]

    SigCom LINCS: 12 data and metadata search engine for a million gene expression signatures.Nucleic Acids Re- search, 50(W1):W697–W709, July 2022

    John Erol Evangelista, Daniel J B Clarke, Zhuorui Xie, Alexander Lachmann, Minji Jeon, Kerwin Chen, Kathleen M Jagodnik, Sherry L Jenkins, Maxim V Kuleshov, Megan L Wo- jciechowicz, Stephan C Schürer, Mario Medvedovic, and Avi Ma’ayan. SigCom LINCS: 12 data and metadata search engine for a million gene expression signatures.Nucleic Acids Re- search, 50(W1...

    work page doi:10.1093/nar/gkac328 2022

  22. [22]

    ToxicoDB: an integrated database to mine and visualize large-scale toxicogenomic datasets.Nucleic Acids Research, 48(W1):W455–W462, July 2020

    Sisira Kadambat Nair, Christopher Eeles, Chantal Ho, Gangesh Beri, Esther Yoo, Denis Tkachuk, Amy Tang, Parwaiz Nijrabi, Petr Smirnov, Heewon Seo, Danyel Jennen, and Benjamin Haibe- Kains. ToxicoDB: an integrated database to mine and visualize large-scale toxicogenomic datasets.Nucleic Acids Research, 48(W1):W455–W462, July 2020. ISSN 0305-1048. doi: 10.1...

    work page doi:10.1093/nar/gkaa390 2020

  23. [23]

    PharmOmics: A species- and tissue-specific drug signature database and gene-network-based drug repositioning tool.iScience, 25(4):104052, April 2022

    Yen-Wei Chen, Graciel Diamante, Jessica Ding, Thien Xuan Nghiem, Jessica Yang, Sung-Min Ha, Peter Cohn, Douglas Arneson, Montgomery Blencowe, Jennifer Garcia, Nima Zaghari, Paul Patel, and Xia Yang. PharmOmics: A species- and tissue-specific drug signature database and gene-network-based drug repositioning tool.iScience, 25(4):104052, April 2022. ISSN 258...

    work page doi:10.1016/j.isci.2022.104052 2022

  24. [24]

    pertdata - A collaborative atlas for perturbational omics datasets

    Sun Sunny, Altana Namsaraeva, Jérémie Kalfon, Lukas Heumos, and Alexander Wolf. pertdata - A collaborative atlas for perturbational omics datasets. Lamin Blog, 2026. URL https: //blog.lamin.ai/pertdata

    2026

  25. [25]

    Green, Ciyue Shen, Torsten Gross, Joseph Min, Samuele Garda, Bo Yuan, Linus J

    Stefan Peidli, Tessa D. Green, Ciyue Shen, Torsten Gross, Joseph Min, Samuele Garda, Bo Yuan, Linus J. Schumacher, Jake P. Taylor-King, Debora S. Marks, Augustin Luna, Nils Blüthgen, and Chris Sander. scPerturb: harmonized single-cell perturbation data.Nature Methods, 21(3):531–540, March 2024. ISSN 1548-7105. doi: 10.1038/s41592-023-02144-y. URL https://...

    work page doi:10.1038/s41592-023-02144-y 2024

  26. [26]

    The MicroArray Quality Control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements.Nature Biotechnology, 24(9):1151–1161, September 2006

    Leming Shi, Leming Shi, Laura H Reid, Wendell D Jones, Richard Shippy, Janet A Warrington, Shawn C Baker, Patrick J Collins, Francoise de Longueville, Ernest S Kawasaki, Kathleen Y Lee, Yuling Luo, Yongming Andrew Sun, James C Willey, Robert A Setterquist, Gavin M Fischer, Weida Tong, Yvonne P Dragan, David J Dix, Felix W Frueh, Federico M Goodsaid, Damir...

    work page doi:10.1038/nbt1239 2006

  27. [27]

    Zhenqiang Su, Paweł P Łabaj, Sheng Li, Jean Thierry-Mieg, Danielle Thierry-Mieg, Wei Shi, Charles Wang, Gary P Schroth, Robert A Setterquist, John F Thompson, Wendell D Jones, Wenzhong Xiao, Weihong Xu, Roderick V Jensen, Reagan Kelly, Joshua Xu, Ana Conesa, Cesare Furlanello, Hanlin Gao, Huixiao Hong, Nadereh Jafari, Stan Letovsky, Yang Liao, Fei Lu, Edw...

    work page doi:10.1038/nbt.2957 2014

  28. [28]

    Pearson, Eser Ayanoglu, Susanne Baumhueter, Keith A

    Brigitte Ganter, Stuart Tugendreich, Cecelia I. Pearson, Eser Ayanoglu, Susanne Baumhueter, Keith A. Bostian, Lindsay Brady, Leslie J. Browne, John T. Calvin, Gwo-Jen Day, Naiomi Breckenridge, Shane Dunlea, Barrett P. Eynon, L. Mike Furness, Joe Ferng, Mark R. Fielden, Susan Y . Fujimoto, Li Gong, Christopher Hu, Radha Idury, Michael S. B. Judo, Kyle L. K...

    work page doi:10.1016/j.jbiotec.2005.03.022 2005

  29. [29]

    Open TG-GATEs: a large-scale toxicogenomics database

    Yoshinobu Igarashi, Noriyuki Nakatsu, Tomoya Yamashita, Atsushi Ono, Yasuo Ohno, Tetsuro Urushidani, and Hiroshi Yamada. Open TG-GATEs: a large-scale toxicogenomics database. Nucleic Acids Research, 43(D1):D921–D927, January 2015. ISSN 0305-1048. doi: 10.1093/ nar/gku955. URLhttps://doi.org/10.1093/nar/gku955

    work page doi:10.1093/nar/gku955 2015

  30. [30]

    Sutherland, Robert A

    Jeffrey J. Sutherland, Robert A. Jolly, Keith M. Goldstein, and James L. Stevens. Assess- ing Concordance of Drug-Induced Transcriptional Response in Rodent Liver and Cultured Hepatocytes.PLOS Computational Biology, 12(3):e1004847, March 2016. ISSN 1553-7358. doi: 10.1371/journal.pcbi.1004847. URL https://journals.plos.org/ploscompbiol/ article?id=10.1371...

    work page doi:10.1371/journal.pcbi.1004847 2016

  31. [31]

    Evaluation of connectivity map shows limited reproducibil- ity in drug repositioning.Scientific Reports, 11(1):17624, September 2021

    Nathaniel Lim and Paul Pavlidis. Evaluation of connectivity map shows limited reproducibil- ity in drug repositioning.Scientific Reports, 11(1):17624, September 2021. ISSN 2045-

    2021

  32. [32]

    URL https://www.nature.com/articles/ s41598-021-97005-z

    doi: 10.1038/s41598-021-97005-z. URL https://www.nature.com/articles/ s41598-021-97005-z

    work page doi:10.1038/s41598-021-97005-z

  33. [33]

    AetherCell: A generative engine for virtual cell perturbation and in vivo drug discovery, March 2026

    Wenyuan Li, Yang Chen, Zhaoyi Peng, Lei Xiang, Dong Wang, and Zhi Xie. AetherCell: A generative engine for virtual cell perturbation and in vivo drug discovery, March 2026. URL https://www.biorxiv.org/content/10.64898/2026.03.13.710968v1. ISSN: 2692- 8205 Pages: 2026.03.13.710968 Section: New Results

    work page doi:10.64898/2026.03.13.710968v1 2026

  34. [34]

    Rosenfeld, Sheng Ding, and Xiang-Dong Fu

    Hairi Li, Hongyan Zhou, Dong Wang, Jinsong Qiu, Yu Zhou, Xiangqiang Li, Michael G. Rosenfeld, Sheng Ding, and Xiang-Dong Fu. Versatile pathway-centric approach based on 14 high-throughput sequencing to anticancer drug discovery.Proceedings of the National Academy of Sciences, 109(12):4609–4614, March 2012. doi: 10.1073/pnas.1200305109. URL https: //www.pn...

    work page doi:10.1073/pnas.1200305109 2012

  35. [35]

    Novartis/drug-seq u2os moabox dataset, March 2025

    Andrea Hadjikyriacou, Chian Yang, Martin Henault, Robin Ge, Leandra Mansur, Alicia Lin- deman, Carsten Russ, Steffen Renner, Marc Hild, Jeremy Jenkins, Caroline Gubser-Keller, Jingyao Li, Daniel J Ho, Marilisa Neri, Frederic D Sigoillot, and Robert Ihry. Novartis/drug-seq u2os moabox dataset, March 2025. URLhttps://doi.org/10.5281/zenodo.14291446

    work page doi:10.5281/zenodo.14291446 2025

  36. [36]

    vcpi-0001

    Virtual Cell Pharmacology Initiative. vcpi-0001. Dataset release on VCPI portal, 2026. URL https://datapoints.ginkgo.bio/. Posted 2026-03-18, accessed 2026-04-12

    2026

  37. [37]

    vcpi-0002

    Virtual Cell Pharmacology Initiative. vcpi-0002. Dataset release on VCPI portal, 2026. URL https://datapoints.ginkgo.bio/. Posted 2026-04-01, accessed 2026-04-12

    2026

  38. [38]

    Mapping the Transcriptional Landscape of Drug Responses in Primary Human Cells Using High-Throughput DRUG-seq, June 2025

    Lauren Baugh, Sébastien Vigneau, Srijani Sridhar, Sarah Boswell, George Pilitsis, John Bradley, Olga Allen, Laura Rand, Amanda Riccardi, Kelsey Roach, Kevin Lema, and Ayla Ergun. Mapping the Transcriptional Landscape of Drug Responses in Primary Human Cells Using High-Throughput DRUG-seq, June 2025. URL https://www.biorxiv.org/content/10. 1101/2025.06.03....

    2025

  39. [39]

    A large- scale human toxicogenomics resource for drug-induced liver injury prediction.Nature Commu- nications, 16(1):9860, November 2025

    V olker Bergen, Konstantia Kodella, Sreenath Srikrishnan, Ornella Barrandon, Sara Anderson, Max Rogers-Grazado, Casey Fowler, Hirit Beyene, Nicole Robichaud, Timothy Fulton, Nina Lapchyk, Mauricio Cortes, Nick Plugis, Matthew Goddeeris, and Mahdi Zamanighomi. A large- scale human toxicogenomics resource for drug-induced liver injury prediction.Nature Comm...

    work page doi:10.1038/s41467-025-65690-3 2025

  40. [40]

    Zhu, E.M

    Y .Y . Zhu, E.M. Machleder, A. Chenchik, R. Li, and P.D. Siebert. Reverse Transcrip- tase Template Switching: A SMART™ Approach for Full-Length cDNA Library Con- struction.BioTechniques, 30(4):892–897, April 2001. ISSN 0736-6205. doi: 10. 2144/01304pf02. URL https://www.tandfonline.com/doi/abs/10.2144/01304pf02. _eprint: https://www.tandfonline.com/doi/pd...

    work page doi:10.2144/01304pf02 2001

  41. [41]

    Mas-Rosario, Rico Meinl, Jalil Nourisa, Jared Tumiel, Tin M

    Artur Szałata, Andrew Benz, Robrecht Cannoodt, Mauricio Cortes, Jason Fong, Sunil Kuppasani, Richard Lieberman, Tianyu Liu, Javier A. Mas-Rosario, Rico Meinl, Jalil Nourisa, Jared Tumiel, Tin M. Tunjic, Mengbo Wang, Noah Weber, Hongyu Zhao, Benedict Anchang, Fabian J. Theis, Malte D. Luecken, and Daniel B. Burkhardt. A benchmark for prediction of transcri...

    work page doi:10.52202/079017-0650 2024

  42. [42]

    Grace X. Y . Zheng, Jessica M. Terry, Phillip Belgrader, Paul Ryvkin, Zachary W. Bent, Ryan Wilson, Solongo B. Ziraldo, Tobias D. Wheeler, Geoff P. McDermott, Junjie Zhu, Mark T. Gregory, Joe Shuga, Luz Montesclaros, Jason G. Underwood, Donald A. Masquelier, Stefanie Y . Nishimura, Michael Schnall-Levin, Paul W. Wyatt, Christopher M. Hindson, Rajiv Bharad...

    work page doi:10.1038/ncomms14049 2017

  43. [43]

    Ensembl 2025.Nucleic Acids Research, 53(D1):D948–D957, January 2025

    Sarah C Dyer, Olanrewaju Austine-Orimoloye, Andrey G Azov, Matthieu Barba, If Barnes, Vianey Paola Barrera-Enriquez, Arne Becker, Ruth Bennett, Martin Beracochea, Andrew Berry, Jyothish Bhai, Simarpreet Kaur Bhurji, Sanjay Boddu, Paulo R Branco Lins, Lucy Brooks, Shashank Budhanuru Ramaraju, Lahcen I Campbell, Manuel Carbajo Martinez, Mehrnaz Charkhchi, L...

    work page doi:10.1093/nar/gkae1071 2025

  44. [44]

    Robinson, Davis J

    Mark D. Robinson, Davis J. McCarthy, and Gordon K. Smyth. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data.Bioinformatics, 26(1): 139–140, January 2010. ISSN 1367-4803. doi: 10.1093/bioinformatics/btp616. URL https: //doi.org/10.1093/bioinformatics/btp616

    work page doi:10.1093/bioinformatics/btp616 2010

  45. [45]

    Taylor-King

    Luka Kovaˇcevi´c, Thomas Gaudelet, James Opzoomer, Hagen Triendl, John Whittaker, Caroline Uhler, Lindsay Edwards, and Jake P. Taylor-King. No Foundations without Foundations – Why semi-mechanistic models are essential for regulatory biology, January 2025. URL http://arxiv.org/abs/2501.19178. arXiv:2501.19178 [cs]

    work page arXiv 2025

  46. [46]

    PubChem 2025 update.Nucleic Acids Research, 53(D1):D1516–D1525, January 2025

    Sunghwan Kim, Jie Chen, Tiejun Cheng, Asta Gindulyte, Jia He, Siqian He, Qingliang Li, Benjamin A Shoemaker, Paul A Thiessen, Bo Yu, Leonid Zaslavsky, Jian Zhang, and Evan E Bolton. PubChem 2025 update.Nucleic Acids Research, 53(D1):D1516–D1525, January 2025. ISSN 0305-1048, 1362-4962. doi: 10.1093/nar/gkae1059. URL https://academic.oup. com/nar/article/5...

    work page doi:10.1093/nar/gkae1059 2025

  47. [47]

    Gdpx1: Gdpx1 functional genomics dataset: Drug-seq + chemical perturba- tion in a549 cells

    Ginkgo Datapoints. Gdpx1: Gdpx1 functional genomics dataset: Drug-seq + chemical perturba- tion in a549 cells. Hugging Face dataset repository, 2025. URL https://huggingface.co/ datasets/ginkgo-datapoints/GDPx1. Accessed 2026-04-12

    2025

  48. [48]

    Gdpx4: Gdpx4 functional genomics dataset

    Ginkgo Datapoints. Gdpx4: Gdpx4 functional genomics dataset. Hugging Face dataset reposi- tory, 2026. URL https://huggingface.co/datasets/ginkgo-datapoints/GDPx4. Ac- cessed 2026-04-12

    2026

  49. [49]

    Stanifer, Gideon Bosker, Xiaoyun Sun, Sergio Triana, Patricio Doldan, Federico La Manna, Marta De Menna, Ronald B

    Pasquale Laise, Megan L. Stanifer, Gideon Bosker, Xiaoyun Sun, Sergio Triana, Patricio Doldan, Federico La Manna, Marta De Menna, Ronald B. Realubit, Sergey Pampou, Charles Karan, Theodore Alexandrov, Marianna Kruithof-de Julio, Andrea Califano, Steeve Boulant, and Mar- iano J. Alvarez. A model for network-based identification and pharmacological targetin...

    work page doi:10.1038/s42003-022-03663-8 2022

  50. [50]

    Gao, Akram Baharlouei, Kyra Thrush-Evensen, Nina Riehs, Amy F

    Joel Dapello, Marcel Nassar, Ridvan Eksi, Ban Wang, Jules Gagnon-Marchand, Kenneth T. Gao, Akram Baharlouei, Kyra Thrush-Evensen, Nina Riehs, Amy F. Peterson, Aniket Tolpadi, Abhejit Rajagopal, Henry E. Miller, Ashley Mae Conard, David Alvarez-Melis, Rory Stark, Simone Bianco, Morgan Levine, Ava P. Amini, Alex Xijie Lu, Nicolo Fusi, Ravi Pandya, Valentina...

    2025

  51. [51]

    A multiplex single-cell RNA-Seq pharmacotranscriptomics pipeline for drug discovery.Nature Chemical Biology, 21(3):432–442, March 2025

    Alice Dini, Harlan Barker, Emilia Piki, Subodh Sharma, Juuli Raivola, Astrid Murumägi, and Daniela Ungureanu. A multiplex single-cell RNA-Seq pharmacotranscriptomics pipeline for drug discovery.Nature Chemical Biology, 21(3):432–442, March 2025. ISSN 1552-

    2025

  52. [52]

    URL https://www.nature.com/articles/ s41589-024-01761-8

    doi: 10.1038/s41589-024-01761-8. URL https://www.nature.com/articles/ s41589-024-01761-8

    work page doi:10.1038/s41589-024-01761-8

  53. [53]

    Rood, Anna Hupalowska, and Aviv Regev

    Jennifer E. Rood, Anna Hupalowska, and Aviv Regev. Toward a foundation model of causal cell and tissue biology with a Perturbation Cell and Tissue Atlas.Cell, 187(17):4520–4545, August 2024. ISSN 0092-8674, 1097-4172. doi: 10.1016/j.cell.2024.07.035. URL https: //www.cell.com/cell/abstract/S0092-8674(24)00829-8. 16

    work page doi:10.1016/j.cell.2024.07.035 2024

  54. [54]

    Clarke and Gary D

    Zoe A. Clarke and Gary D. Bader. MALAT1 expression indicates cell quality in single-cell RNA sequencing data, July 2024. URL https://www.biorxiv.org/content/10.1101/ 2024.07.14.603469v2. Pages: 2024.07.14.603469 Section: New Results

    2024

  55. [55]

    decoupleR: ensemble of computational methods to infer biological activities from omics data.Bioinformatics Advances, 2(1):vbac016, January

    Pau Badia-i Mompel, Jesús Vélez Santiago, Jana Braunger, Celina Geiss, Daniel Dim- itrov, Sophia Müller-Dott, Petr Taus, Aurelien Dugourd, Christian H Holland, Ricardo O Ramirez Flores, and Julio Saez-Rodriguez. decoupleR: ensemble of computational methods to infer biological activities from omics data.Bioinformatics Advances, 2(1):vbac016, January

  56. [56]

    doi: 10.1093/bioadv/vbac016

    ISSN 2635-0041. doi: 10.1093/bioadv/vbac016. URL https://academic.oup.com/ bioinformaticsadvances/article/doi/10.1093/bioadv/vbac016/6544613

    work page doi:10.1093/bioadv/vbac016

  57. [57]

    Theis, Philipp Angerer, and F

    Isaac Virshup, Sergei Rybakov, Fabian J. Theis, Philipp Angerer, and F. Alexander Wolf. anndata: Access and store annotated data matrices.Journal of Open Source Software, 9 (101):4371, September 2024. ISSN 2475-9066. doi: 10.21105/joss.04371. URL https: //joss.theoj.org/papers/10.21105/joss.04371

    work page doi:10.21105/joss.04371 2024

  58. [58]

    The Cellosaurus, a Cell-Line Knowledge Resource.Journal of Biomolecular Techniques : JBT, 29(2):25–38, July 2018

    Amos Bairoch. The Cellosaurus, a Cell-Line Knowledge Resource.Journal of Biomolecular Techniques : JBT, 29(2):25–38, July 2018. ISSN 1524-0215. doi: 10.7171/jbt.18-2902-002. URLhttps://pmc.ncbi.nlm.nih.gov/articles/PMC5945021/

    work page doi:10.7171/jbt.18-2902-002 2018

  59. [59]

    Diehl, Terrence F

    Alexander D. Diehl, Terrence F. Meehan, Yvonne M. Bradford, Matthew H. Brush, Wasila M. Dahdul, David S. Dougall, Yongqun He, David Osumi-Sutherland, Alan Ruttenberg, Sirarat Sarntivijai, Ceri E. Van Slyke, Nicole A. Vasilevsky, Melissa A. Haendel, Judith A. Blake, and Christopher J. Mungall. The Cell Ontology 2016: enhanced content, modularization, and o...

    work page doi:10.1186/s13326-016-0088-7 2016

  60. [60]

    Disentanglement of single-cell data with biolord.Nature Biotechnology, 42(11):1678–1683, November 2024

    Zoe Piran, Niv Cohen, Yedid Hoshen, and Mor Nitzan. Disentanglement of single-cell data with biolord.Nature Biotechnology, 42(11):1678–1683, November 2024. ISSN 1546-

    2024

  61. [61]

    learn its own embedding

    doi: 10.1038/s41587-023-02079-x. URL https://www.nature.com/articles/ s41587-023-02079-x. 17 A Formal problem setup We study perturbation effects derived from bulk or pseudobulk transcriptomic measurements. A perturbation condition is indexed by ξ= (a, p, c, d, t)∈ A × P × C × D × T, where a denotes dataset or assay domain, p perturbation identity, c cell...

    work page doi:10.1038/s41587-023-02079-x 2023