pith. sign in

arxiv: 2605.29008 · v1 · pith:VVWSIWXSnew · submitted 2026-05-27 · 💻 cs.LG

Causal Intelligence for Constraint-Aware Intervention Design to Induce State Transitions

Zixuan Song , Uwe Mueller , Dimitris V. Manatakis This is my paper

Pith reviewed 2026-06-29 14:19 UTC · model grok-4.3

classification 💻 cs.LG
keywords causal discoveryintervention designstate transitionsstructural causal modelsconstraint-aware optimizationsynthetic benchmarksbiological datasets
0
0 comments X

The pith

COAST learns causal graphs from source and target data to optimize constrained interventions that induce desired state transitions.

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

The paper presents COAST as a framework that takes data from an initial source state and a desired target state, learns context-specific causal graphs and structural causal models, and attributes observed shifts between the states to specific causal mechanisms. It then applies a new constraint-aware multi-objective optimization that selects interventions while trading off how effectively they produce the transition, how many targets they affect, and how stable the final state remains. The approach is built to be modular so that different causal discovery or modeling tools can be swapped in. This matters for fields where black-box predictors give no guidance on which variables to change experimentally. Tests on synthetic benchmarks and real biological data show it recovers key drivers and produces both single-target and multi-target strategies with explicit mechanistic explanations.

Core claim

Given data characterizing source and target states, COAST learns context-specific causal graphs and structural causal models, attributes observed distributional shifts to mechanism-level causal drivers, and introduces a novel constraint-aware multi-objective optimization formulation that balances transition efficacy, intervention complexity, and target-state stability. The approach is modular and domain-agnostic, integrating feature selection, causal discovery, causal modeling, and intervention identification and evaluation through interchangeable components.

What carries the argument

The constraint-aware multi-objective optimization formulation that balances transition efficacy, intervention complexity, and target-state stability on top of learned structural causal models.

If this is right

  • COAST recovers key causal drivers from the data.
  • It identifies both single-target and multi-target intervention strategies.
  • The selected strategies come with explicit mechanistic rationales for experimental follow-up.
  • The framework remains modular so that different causal-discovery algorithms can be substituted without changing the rest of the pipeline.
  • It works across both synthetic benchmarks and real biological datasets.

Where Pith is reading between the lines

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

  • The same pipeline could be tested on non-biological systems such as economic or climate networks where state shifts are also desired.
  • Adding a step that incorporates limited interventional data might improve the accuracy of the recovered graphs beyond what purely observational data allows.
  • The optimization objective could be extended to time-series data to handle transitions that unfold over multiple steps rather than instantaneous changes.
  • The emphasis on attributing distributional shifts to specific mechanisms suggests that similar attribution steps could improve intervention design even in non-causal predictive models.

Load-bearing premise

The method assumes that observational data from the source and target states alone is enough for causal discovery to recover the correct causal links and to correctly explain the observed differences between the states.

What would settle it

A controlled synthetic dataset with a known ground-truth causal graph in which the interventions selected by COAST fail to produce the target state when the learned graph is used to simulate the interventions.

read the original abstract

Driving a system from one state to another through targeted interventions is a fundamental challenge in science, yet most predictive models offer limited mechanistic insight and no principled framework for decision-making. Here we present COAST (Causally Optimal Actions for State Transitions), a causal-intelligence approach for the in-silico design of constrained interventions that induce user-defined state transitions. Given data characterizing source and target states, COAST learns context-specific causal graphs and structural causal models, attributes observed distributional shifts to mechanism-level causal drivers, and introduces a novel constraint-aware multi-objective optimization formulation that balances transition efficacy, intervention complexity, and target-state stability. The approach is modular and domain-agnostic, integrating feature selection, causal discovery, causal modeling, and intervention identification and evaluation through interchangeable components. Across synthetic benchmarks and real biological datasets, COAST recovers key causal drivers and identifies robust single- and multi-target intervention strategies that achieve desired state transitions, accompanied by transparent mechanistic rationales to guide experimental validation.

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 / 2 minor

Summary. The manuscript introduces COAST, a modular, domain-agnostic framework that takes observational data on source and target states, applies causal discovery to learn context-specific graphs and structural causal models, attributes distributional shifts to mechanism-level drivers, and solves a novel constraint-aware multi-objective optimization to identify single- and multi-target interventions that achieve desired state transitions while balancing efficacy, complexity, and post-intervention stability. Claims of recovering key causal drivers and producing robust, mechanistically interpretable strategies are supported by results on synthetic benchmarks and real biological datasets.

Significance. If the causal-recovery step is reliable, the work could meaningfully advance intervention design in systems biology and related fields by supplying both actionable strategies and transparent mechanistic rationales. The modular architecture and explicit multi-objective formulation are constructive contributions; the emphasis on constraint awareness and stability is practically relevant. The absence of new identifiability results or extensive robustness checks against common biological-data pathologies, however, limits the strength of the significance claim.

major comments (2)
  1. [Causal discovery and SCM learning component (method description)] The central claim that COAST recovers key causal drivers and identifies robust interventions rests on the premise that standard causal-discovery algorithms applied to observational samples from source and target states produce faithful context-specific DAGs and SCMs. The manuscript provides no new identifiability results or systematic sensitivity analyses addressing latent confounders, non-stationarity between states, or finite-sample issues that routinely arise in biological data; without such evidence the attribution of distributional shifts and the subsequent optimization outputs remain vulnerable to misspecification.
  2. [Intervention identification and evaluation] The constraint-aware multi-objective optimization is presented as balancing transition efficacy, intervention complexity, and target-state stability, yet the manuscript does not detail how the learned SCMs are used to evaluate post-intervention stability or how the attribution of mechanism-level drivers is validated against ground-truth interventions in the synthetic benchmarks. This gap makes it difficult to assess whether the reported mechanistic rationales are robust or merely post-hoc interpretations.
minor comments (2)
  1. The abstract and method overview would benefit from a concise statement of the precise causal-discovery algorithms and SCM estimators employed, together with any hyper-parameter choices that affect graph recovery.
  2. Figure captions and table legends should explicitly state the number of independent runs, the exact metrics used for “recovery of key causal drivers,” and whether error bars reflect variability across data splits or random seeds.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. We address each major comment below, clarifying the scope of our contributions and indicating revisions to strengthen the manuscript where appropriate.

read point-by-point responses

  1. Referee: [Causal discovery and SCM learning component (method description)] The central claim that COAST recovers key causal drivers and identifies robust interventions rests on the premise that standard causal-discovery algorithms applied to observational samples from source and target states produce faithful context-specific DAGs and SCMs. The manuscript provides no new identifiability results or systematic sensitivity analyses addressing latent confounders, non-stationarity between states, or finite-sample issues that routinely arise in biological data; without such evidence the attribution of distributional shifts and the subsequent optimization outputs remain vulnerable to misspecification.

    Authors: We acknowledge that the manuscript does not provide new theoretical identifiability results, as the primary contribution lies in the modular end-to-end framework for constraint-aware intervention design rather than advances in causal discovery theory. The approach is explicitly designed to integrate interchangeable existing causal discovery methods. To address concerns about robustness, we will add systematic sensitivity analyses in the revised manuscript, including evaluations under latent confounding, non-stationarity, and varying sample sizes on both synthetic and biological data. revision: partial

  2. Referee: [Intervention identification and evaluation] The constraint-aware multi-objective optimization is presented as balancing transition efficacy, intervention complexity, and target-state stability, yet the manuscript does not detail how the learned SCMs are used to evaluate post-intervention stability or how the attribution of mechanism-level drivers is validated against ground-truth interventions in the synthetic benchmarks. This gap makes it difficult to assess whether the reported mechanistic rationales are robust or merely post-hoc interpretations.

    Authors: We will expand the methods and experimental sections to explicitly describe the simulation procedure using the learned SCMs for evaluating post-intervention stability (via forward simulation of the intervened system and computation of distributional metrics relative to the target state). We will also include direct validation of the mechanism-level attribution step against ground-truth causal drivers in the synthetic benchmarks, reporting quantitative agreement metrics to demonstrate that the rationales are not post-hoc. revision: yes

Circularity Check

0 steps flagged

No circularity: method learns from data without self-referential reductions

full rationale

The abstract and method description frame COAST as integrating standard causal discovery, modeling, and optimization components applied to observational source/target data. No equations or claims are presented in which a 'prediction' or 'result' is shown to equal its own fitted inputs by construction, nor is any load-bearing premise justified solely via self-citation. The central workflow (learn graphs from data, attribute shifts, optimize interventions) remains externally falsifiable against held-out data and does not reduce to renaming or self-definition. This is the expected non-finding for a modular applied pipeline.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard causal inference assumptions that are not detailed in the abstract; no free parameters or invented entities are mentioned.

axioms (2)
  • domain assumption Causal discovery from observational data recovers the true context-specific causal graph
    Invoked when the method learns causal graphs to attribute shifts to drivers.
  • domain assumption Structural causal models can be fitted to explain observed distributional shifts between states
    Used to attribute shifts to mechanism-level causal drivers.

pith-pipeline@v0.9.1-grok · 5702 in / 1225 out tokens · 37488 ms · 2026-06-29T14:19:25.893219+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

65 extracted references · 12 canonical work pages

  1. [1]

    Early-warning signals for critical transitions,

    Scheffer, M., Bascompte, J., Brock, W. A.et al.Early-warning signals for critical transitions.Nature 461, 53–59 (2009). doi:10.1038/nature08227

    work page doi:10.1038/nature08227 2009

  2. [2]

    & Costello, J

    Correia, C., Ung, C.-Y., Li, H. & Costello, J. C. State-of-the-art hypothesis-driven systems pharmacology and artificial intelligence approaches to decipher disease complexity.Front. Pharmacol.16, 1593164 (2025). doi:10.3389/fphar.2025.1593164

    work page doi:10.3389/fphar.2025.1593164 2025

  3. [3]

    Multitissue multiomics systems biology to dissect complex diseases.Trends Mol

    Yang, X. Multitissue multiomics systems biology to dissect complex diseases.Trends Mol. Med.26, 718–728 (2020). doi:10.1016/j.molmed.2020.04.006

    work page doi:10.1016/j.molmed.2020.04.006 2020

  4. [4]

    & Sarkar, R

    Panditrao, G., Bhowmick, R., Meena, C. & Sarkar, R. R. Emerging landscape of molecular interaction networks: opportunities, challenges and prospects.J. Biosci.47, 24 (2022). doi:10.1007/s12038-022- 00253-y 19

    work page doi:10.1007/s12038-022- 2022

  5. [5]

    Pearl, J.Causality: Models, Reasoning, and Inference. 2nd edn. Cambridge Univ. Press (2009)

    2009

  6. [6]

    Toward causal representation learning,

    Sch¨ olkopf, B., Locatello, F., Bauer, S., Ke, N. R., Kalchbrenner, N., Goyal, A. & Bengio, Y. Toward causal representation learning.Proc. IEEE109, 612–634 (2021). doi:10.1109/JPROC.2021.3058954

    work page doi:10.1109/jproc.2021.3058954 2021

  7. [7]

    Lotfollahi, M., Wolf, F. A. & Theis, F. J. scGen predicts single-cell perturbation responses.Nat. Methods 16, 715–721 (2019). doi:10.1038/s41592-019-0494-8

    work page doi:10.1038/s41592-019-0494-8 2019

  8. [8]

    & Others Predicting cellular responses to complex perturbations in high-throughput screens.Molecular Systems Biology.19, MSB202211517 (2023)

    Lotfollahi, M., Klimovskaia Susmelj, A., De Donno, C., Hetzel, L., Ji, Y., Ibarra, I., Srivatsan, S., Naghipourfar, M., Daza, R., Martin, B. & Others Predicting cellular responses to complex perturbations in high-throughput screens.Molecular Systems Biology.19, MSB202211517 (2023)

    2023

  9. [9]

    & Theis, F

    Hetzel, L., B¨ ohm, S., Kilbertus, N., G¨ unnemann, S., Lotfollahi, M. & Theis, F. J. Predicting cellular responses to novel drug perturbations at single-cell resolution.Adv. Neural Inf. Process. Syst.35(2022)

    2022

  10. [10]

    & Leskovec, J

    Roohani, Y., Huang, K. & Leskovec, J. Predicting transcriptional outcomes of novel multigene perturbations with GEARS.Nature Biotechnology.42, 927-935 (2024)

    2024

  11. [11]

    Stark, Gabriele Gut, Jacobo Sarabia del Castillo, Mitch Levesque, Kjong- Van Lehmann, Lucas Pelkmans, Andreas Krause, and Gunnar R ¨atsch

    Bunne, C., Stark, S. G., Gut, G., Sarabia del Castillo, J., Levesque, M., Lehmann, K.-V., Pelkmans, L., Krause, A. & R¨ atsch, G. Learning single-cell perturbation responses using neural optimal transport. Nat. Methods20, 1759-1768 (2023). doi:10.1038/s41592-023-01969-x

    work page doi:10.1038/s41592-023-01969-x 2023

  12. [12]

    Gonzalez, G., Lin, X., Herath, I.et al.Combinatorial prediction of therapeutic perturbations using causally inspired neural networks.Nat. Biomed. Eng.(2025). doi:10.1038/s41551-025-01481-x

    work page doi:10.1038/s41551-025-01481-x 2025

  13. [13]

    & Chen, Z

    Zhu, S., Ng, I. & Chen, Z. Causal discovery with reinforcement learning.International Conference On Learning Representations(2020)

    2020

  14. [14]

    & Uhler, C

    Zhang, J., Cammarata, L., Squires, C., Sapsis, T. & Uhler, C. Active learning for optimal intervention design in causal models.Nature Machine Intelligence.5, 1066-1075 (2023)

    2023

  15. [15]

    Frangieh, C., Melms, J., Thakore,et al.Multimodal pooled Perturb-CITE-seq screens in patient models define mechanisms of cancer immune evasion.Nature Genetics.53, 332-341 (2021)

    2021

  16. [16]

    Single-cell RNA-Seq reveals hypothalamic cell diversity

    Chen, R., Wu, X., Jiang, L., and Zhang, Y. Single-cell RNA-Seq reveals hypothalamic cell diversity. Cell Reports,18(13), 3227–3241 (2017)

    2017

  17. [17]

    D., Hanson, S

    Ramsey, J. D., Hanson, S. J., Hanson, C., Halchenko, Y. O., Poldrack, R. A. & Glymour, C. Six problems for causal inference from fMRI.NeuroImage49, 1545–1558 (2010)

    2010

  18. [18]

    Marques, S., Zeisel, A., Codeluppi, S., van Bruggen, D.,et al.Oligodendrocyte heterogeneity in the mouse juvenile and adult central nervous system.Science,352(6291), 1326–1329 (2016)

    2016

  19. [19]

    Claudin-11/OSP-based tight junctions of myelin sheaths in brain and Sertoli cells in testis.Journal of Cell Biology,145(3), 579–588 (1999)

    Morita, K., Sasaki, H., Fujimoto, K., Furuse, M., and Tsukita, S. Claudin-11/OSP-based tight junctions of myelin sheaths in brain and Sertoli cells in testis.Journal of Cell Biology,145(3), 579–588 (1999)

    1999

  20. [20]

    and Menzies, S.L

    Connor, J.R. and Menzies, S.L. Relationship of iron to oligodendrocytes and myelination.Glia,17(2), 83–93 (1996)

    1996

  21. [21]

    Ortiz, E., Pasquini, J.M., Thompson, K., Felt, B., Butkus, G., Beard, J., and Connor, J.R. Effect of manipulation of iron storage, transport, or availability on myelin composition and brain iron content in three different animal models.Journal of Neuroscience Research,77(5), 681–689 (2004)

    2004

  22. [22]

    Apolipoprotein D is involved in the mechanisms regulating protection from oxidative stress.Aging Cell,7(4), 506–515 (2008)

    Ganfornina, M.D., Do Carmo, S., Lora, J.M., Torres-Schumann, S., Vogel, M., Allhorn, M., Gonz´ alez, C., Bastiani, M.J., Rassart, E., and Sanchez, D. Apolipoprotein D is involved in the mechanisms regulating protection from oxidative stress.Aging Cell,7(4), 506–515 (2008)

    2008

  23. [23]

    and Pham-Dinh, D

    Baumann, N. and Pham-Dinh, D. Biology of oligodendrocyte and myelin in the mammalian central nervous system.Physiological Reviews,81(2), 871–927 (2001). 20

    2001

  24. [24]

    and Werner, H.B

    Nave, K.A. and Werner, H.B. Myelination of the nervous system: mechanisms and functions.Annual Review of Cell and Developmental Biology,30, 503–533 (2014)

    2014

  25. [25]

    Ermin, a myelinating oligodendrocyte- specific protein that regulates cell morphology.Journal of Neuroscience,26(3), 757–762 (2006)

    Brockschnieder, D., Sabanay, H., Riethmacher, D., and Peles, E. Ermin, a myelinating oligodendrocyte- specific protein that regulates cell morphology.Journal of Neuroscience,26(3), 757–762 (2006)

    2006

  26. [26]

    Identification of Tmem10/Opalin as an oligodendrocyte enriched gene using expression profiling combined with genetic cell ablation.Glia,56(11), 1176–1186 (2008)

    Golan, N., Adamsky, K., Kartvelishvily, E., Brockschnieder, D., M¨ obius, W., Spiegel, I., Roth, A.D., Thomson, C.E., Rechavi, G., and Peles, E. Identification of Tmem10/Opalin as an oligodendrocyte enriched gene using expression profiling combined with genetic cell ablation.Glia,56(11), 1176–1186 (2008)

    2008

  27. [27]

    & Peterson, H

    Kolberg, L., Raudvere, U., Kuzmin, I., Adler, P., Vilo, J. & Peterson, H. g:Profiler—interoperable web service for functional enrichment analysis and gene identifier mapping (2023 update).Nucleic Acids Res. 51(W1), W207–W212 (2023). https://doi.org/10.1093/nar/gkad347

    work page doi:10.1093/nar/gkad347 2023

  28. [28]

    and Lu, Q.R

    Emery, B. and Lu, Q.R. Transcriptional and epigenetic regulation of oligodendrocyte development and myelination in the central nervous system.Cold Spring Harbor Perspectives in Biology,7(9), a020461 (2015)

    2015

  29. [29]

    & Elisseeff, A

    Guyon, I. & Elisseeff, A. An introduction to variable and feature selection.J. Mach. Learn. Res.3, 1157–1182 (2003)

    2003

  30. [30]

    Fan, J. & Lv, J. A selective overview of variable selection in high-dimensional feature space.Stat. Sin. 20, 101–148 (2010)

    2010

  31. [31]

    & B¨ uhlmann, P

    Meinshausen, N. & B¨ uhlmann, P. High-dimensional graphs and variable selection with the lasso.Ann. Stat.34, 1436–1462 (2006)

    2006

  32. [32]

    & Sch¨ olkopf, B.Elements of Causal Inference: Foundations and Learning Algorithms

    Peters, J., Janzing, D. & Sch¨ olkopf, B.Elements of Causal Inference: Foundations and Learning Algorithms. MIT Press (2017)

    2017

  33. [33]

    J.Handbook of Parametric and Nonparametric Statistical Procedures

    Sheskin, D. J.Handbook of Parametric and Nonparametric Statistical Procedures. 5th edn. Chapman & Hall/CRC (2011)

    2011

  34. [34]

    I., Huber, W

    Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2.Genome Biol.15, 550 (2014)

    2014

  35. [35]

    D., McCarthy, D

    Robinson, M. D., McCarthy, D. J. & Smyth, G. K. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data.Bioinformatics26, 139–140 (2010)

    2010

  36. [36]

    W., Chen, Y., Shi, W

    Law, C. W., Chen, Y., Shi, W. & Smyth, G. K. voom: Precision weights unlock linear model analysis tools for RNA-seq read counts.Genome Biol.15, R29 (2014)

    2014

  37. [37]

    A., Irrthum, A., Wehenkel, L

    Huynh-Thu, V. A., Irrthum, A., Wehenkel, L. & Geurts, P. Inferring regulatory networks from expression data using tree-based methods.PLoS One5, e12776 (2010)

    2010

  38. [38]

    & Hastie, T

    Zou, H. & Hastie, T. Regularization and variable selection via the elastic net.J. R. Stat. Soc. B67, 301–320 (2005)

    2005

  39. [39]

    & B¨ uhlmann, P

    Meinshausen, N. & B¨ uhlmann, P. Stability selection.J. R. Stat. Soc. B72, 417–473 (2010)

    2010

  40. [40]

    Cover, T. M. & Thomas, J. A.Elements of Information Theory. 2nd edn. Wiley (2006)

    2006

  41. [41]

    M., Sprecher, E., Trifonov, V

    Yu, H., Kim, P. M., Sprecher, E., Trifonov, V. & Gerstein, M. The importance of bottlenecks in protein networks: correlation with gene essentiality and expression dynamics.PLoS Comput. Biol.3, e59 (2007)

    2007

  42. [42]

    Dissection of hubs and bottlenecks in a protein- protein interaction network.Computational Biology and Chemistry, 102:107802 (2023)

    Nithya C., Kiran M., Nagarajaram H.A. Dissection of hubs and bottlenecks in a protein- protein interaction network.Computational Biology and Chemistry, 102:107802 (2023). doi: 10.1016/j.compbiolchem.2022.107802

    work page doi:10.1016/j.compbiolchem.2022.107802 2023

  43. [43]

    Freeman, L. C. A set of measures of centrality based on betweenness.Sociometry40, 35–41 (1977). 21

    1977

  44. [44]

    Review of Causal Discovery Methods Based on Graphical Models,

    Glymour, C., Zhang, K. & Spirtes, P. Review of causal discovery methods based on graphical models. Front. Genet.10, 524 (2019). doi:10.3389/fgene.2019.00524

    work page doi:10.3389/fgene.2019.00524 2019

  45. [45]

    & Scheines, R.Causation, Prediction, and Search

    Spirtes, P., Glymour, C. & Scheines, R.Causation, Prediction, and Search. 2nd edn. MIT Press (2000)

    2000

  46. [46]

    Chickering, D. M. Optimal structure identification with greedy search.J. Mach. Learn. Res.3, 507–554 (2002). doi:10.1162/153244303321897717

    work page doi:10.1162/153244303321897717 2002

  47. [47]

    & Xing, E

    Zheng, X., Aragam, B., Ravikumar, P. & Xing, E. P. DAGs with NO TEARS: continuous optimization for structure learning.Adv. Neural Inf. Process. Syst.31, 9472–9483 (2018)

    2018

  48. [48]

    O., Hyv¨ arinen, A

    Shimizu, S., Hoyer, P. O., Hyv¨ arinen, A. & Kerminen, A. A linear non-Gaussian acyclic model for causal discovery.J. Mach. Learn. Res.7, 2003–2030 (2006)

    2003

  49. [49]

    & B¨ uhlmann, P

    Hauser, A. & B¨ uhlmann, P. Characterization and greedy learning of interventional Markov equivalence classes of directed acyclic graphs.J. Mach. Learn. Res.13, 2409–2464 (2012)

    2012

  50. [50]

    J. D. Ramsey, K. Zhang, M. Glymour, R. Sanchez-Romero, B. Huang, I. Ebert-Uphoff, S. Samarasinghe, E. A. Barnes, and C. Glymour, TETRAD—A Toolbox for Causal Discovery,in Proceedings of the 8th International Workshop on Climate Informatics, Sept. 19-21, 2018

    2018

  51. [51]

    & Uhler, C

    Wang, Y., Segarra, S. & Uhler, C. High-dimensional joint estimation of multiple directed Gaussian graphical models.Electron. J. Stat.14, 2439–2483 (2020)

    2020

  52. [52]

    https://www.pywhy.org/dowhy/ (accessed 2026)

    Sharma, A.et al.DoWhy: a Python library for causal inference. https://www.pywhy.org/dowhy/ (accessed 2026)

    2026

  53. [53]

    Budhathoki, K., Janzing, D., Bl¨ obaum, P. & Ng, H. Why did the distribution change? InProc. 24th Int. Conf. Artificial Intelligence and Statistics (AISTATS). PMLR130, 1666–1674 (2021)

    2021

  54. [54]

    Shapley, L. S. A value for n-person games. In Kuhn, H. W. & Tucker, A. W. (eds)Contributions to the Theory of Games, Volume II. Annals of Mathematics Studies28, 307–317 (Princeton Univ. Press, 1953)

    1953

  55. [55]

    Sen, R., Shanmugam, K., Dimakis, A. G. & Shakkottai, S. Identifying best interventions through online importance sampling. InProc. 34th Int. Conf. Machine Learning (ICML)3057–3066 (2017)

    2017

  56. [56]

    & Uhler, C

    Zhang, J., Squires, C. & Uhler, C. Matching a desired causal state via shift interventions.Adv. Neural Inf. Process. Syst.34(2021)

    2021

  57. [57]

    & Scheines, R

    Eberhardt, F. & Scheines, R. Interventions and causal inference.Philos. Sci.74, 981–995 (2007)

    2007

  58. [58]

    & Vandenberghe, L.Convex Optimization

    Boyd, S. & Vandenberghe, L.Convex Optimization. Cambridge Univ. Press (2004)

    2004

  59. [59]

    Regression shrinkage and selection via the lasso.J

    Tibshirani, R. Regression shrinkage and selection via the lasso.J. R. Stat. Soc. B58, 267–288 (1996)

    1996

  60. [60]

    & Tibshirani, R

    Friedman, J., Hastie, T. & Tibshirani, R. Regularization paths for generalized linear models via coordinate descent.J. Stat. Softw.33, 1–22 (2010)

    2010

  61. [61]

    The adaptive lasso and its oracle properties.J

    Zou, H. The adaptive lasso and its oracle properties.J. Am. Stat. Assoc.101, 1418–1429 (2006)

    2006

  62. [62]

    Genet.25, 25–29 (2000)

    Ashburner, M.et al.Gene ontology: tool for the unification of biology.Nat. Genet.25, 25–29 (2000)

    2000

  63. [63]

    & Goto, S

    Kanehisa, M. & Goto, S. KEGG: Kyoto Encyclopedia of Genes and Genomes.Nucleic Acids Res.28, 27–30 (2000)

    2000

  64. [64]

    Natl Acad

    Subramanian, A.et al.Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles.Proc. Natl Acad. Sci. USA102, 15545–15550 (2005)

    2005

  65. [65]

    Nucleic Acids Res.53, D1467–D1475 (2025)

    Buniello, A.et al.Open Targets Platform: facilitating therapeutic hypothesis building in drug discovery. Nucleic Acids Res.53, D1467–D1475 (2025). 22 Appendix A Average transition percentage As COAST identifies a series of solutions with different numbers of intervention targets using different values of the regularization strengthλ, average transition pe...

    2025