Perturbation Data for Causal and Disease-Relevant Foundation Models in Biology

Biological foundation models are increasingly trained on large observational datasets such as single-cell transcriptomic atlases, yet most do not explicitly leverage perturbation data. This limits their ability to learn directionality, mechanism, and intervention-relevant structure. In contrast with human language, where syntax and usage often preserve causal and semantic relationships, biological measurements are noisy, sparse, and heavily confounded by cell state, batch, and experimental design. Perturbation experiments, including genetic and chemical interventions, provide a principled way to expose functional dependencies and context-specific responses. This Research Topic will highlight mathematical, statistical, and computational advances that use perturbation data to build more causal, robust, and disease-relevant foundation models for biology, with particular emphasis on human disease, therapeutic discovery, and mechanistic interpretation.A central goal is to move beyond descriptive models that mainly capture correlation toward models that better encode biological mechanisms, predict responses to intervention, and generalize across context. We encourage theoretical, methodological, and applied work that evaluates whether perturbation-informed models improve interpretability, disease relevance, target discovery, and prediction of therapeutic response. Submissions should make a clear mathematical, statistical, or computational contribution while remaining grounded in biologically meaningful questions.This Research Topic seeks contributions that develop rigorous quantitative methods for building biological foundation models from perturbation data, alone or in combination with observational data. We welcome studies on:- causal representation learning;- intervention-aware embeddings;- single-cell perturbation modeling;- counterfactual inference;- dynamical systems approaches;- uncertainty quantification;- benchmark design;- transfer learning across platforms, tissues, and disease settings.We invite Original Research, Reviews, Methods, Perspectives, and related article types that fit Frontiers requirements and advance the mathematics, statistics, or computation of perturbation-informed biological modeling. Submissions should be written for a broad quantitative audience spanning mathematical biology, applied statistics, and data science. Authors are encouraged to state the methodological novelty, assumptions, validation strategy, and biological significance of their work clearly. Contributions may focus on theory, algorithms, simulation, benchmarking, or disease applications, but each submission should demonstrate clear relevance to perturbation-based modeling and its potential to improve disease-focused biological foundation models. Topic editors: Qian Du, Weilong Yang, Yinglun Zhan Source: https://www.frontiersin.org/research-topics/80984/perturbation-data-for-causal-and-disease-relevant-foundation-models-in-biology