MAIGE Group: Mechanistic AI in Genomics
We are a computational genomics group particularly interested in the application of AI and ML in a way that respects biological mechanisms and is both mathematically and statistically sound. We work with large-scale cohort data (population genetics, electronic health records) and cellular assays (single-cell omics and CRISPR perturbation assays).
Our lab's goal is to gain clear biological insights from this complex, noisy and incomplete data. That's why our scientific motto is:
Train smarter models, not larger ones.
To do that, our team draws on our diverse backgrounds to analyze and model data, exploiting principles from biology, mathematics, statistics, and computational science.
We emphasize scientific rigor and excellence, but just as importantly a positive and nurturing team spirit. For us, that means kindness and mutual respect. It also means an open, collaborative attitude both within our lab and towards our fellow scientists in other groups.
Research Topics
We are interested in genetic variant effects on human traits, for example, disease risk, but additionally molecular traits such as gene expression. In particular, we develop methods sensitive enough to detect effects that, while subtle, are critical for unraveling biological mechanisms. This includes effects of rare variants, which are found in few individuals but play an outsized role in disease. These are difficult to study using classical statistics, but they lend themselves to modeling using AI models trained on genomic and protein sequences. It also includes trans-eQTLs, which can be used to understand gene–gene regulatory relationships and how they are disrupted in disease. Our methods aim to fully exploit the single-cell nature scRNA-seq data while also respecting biological principles such as shared regulation of pathways.
Related publications
Vagiaki, D., Heinen, T., Saraswat, M., Clarke, B. & Stegle, O., 2026. Mapping trans-eQTLs at single-cell resolution using Latent Interaction Variational Inference. bioRxiv. DOI: 10.64898/2026.02.04.703363
Clarke, B., Holtkamp, E., Öztürk, H., Mück, M., Wahlberg, M., Meyer, K., Munzlinger, F., Brechtmann, F., Hölzlwimmer, F. R., Lindner, J., Chen, Z., Gagneur, J. & Stegle, O., 2024. Integration of variant annotations using deep set networks boosts rare variant association testing. Nat Genet. DOI: 10.1038/s41588-024-01919-z
Our group builds causal inference methods for reconstructing gene regulatory networks from interventional data, especially CRISPR Perturb-seq screens. For the challenging task of teasing apart direct causal effects from noisy total causal effect measurements, we use mathematically inspired techniques such as parameter learning for dynamical systems. Our current focus lies in combining gene regulatory network inference with population genetics and other modalities in order to better understand regulatory mechanisms behind traits with complex genetic architectures. This includes, for example the rare variant genetic effects on traits and trans-eQTL effects on gene regulation described above. We are also interested in sequence-to-function models of gene regulation, and in using Perturb-seq screens in organoids to model gene regulation in a more realistic biological context.
Related publications
Rohbeck, M., Clarke, B., Mikulik, K., Pettet, A., Stegle, O. & Ueltzhöffer, K., 2024. Bicycle: Intervention-Based Causal Discovery with Cycles, in: Proceedings of the Third Conference on Causal Learning and Reasoning, PMLR. https://proceedings.mlr.press/v236/rohbeck24a.html
Open positions
Recruiting for PhD positions is done through the International PhD Programs of DKFZ and EMBL.
Recruiting for postdocs is through DKFZ Postdoctoral Fellowships and by reaching out directly to us.
For inquiries about bachelor's/master's internships and theses, please send your CV, desired timeframe, and a brief (1-2 paragraph) motivation directly to us.
In case of general questions or to check whether we're participating in a given round of PhD/postdoc recruitment, please use the “Write a message” link next to Brian Clarke's profile.
Get in touch with us
