More details on the model and discovery examples are included in our preprint (link above)!
If you wish to try out LIVI: github.com/PMBio/LIVI
Suggestions and questions are very welcome!
Last but not least, a HUGE thank you to all coauthors and colleagues for their contributions! 🙌
12/12
As scRNA-seq cohorts grow in scale (e.g. FinnGen, TenK10K), so does the need for methods that can handle millions of cells from thousands of donors. LIVI's scalable design enables systematic trans-eQTL discovery in such datasets, revealing cellular mechanisms underlying complex disease risk. 11/n
Beyond single variant effects, LIVI can detect cells and genes influenced by polygenic disease risk. Example: PRSs for rheumatoid arthritis (RA) and celiac disease (CeD), diseases driven by autoantibody production, were linked to immunoglobulin-mediated immune response genes in B cells. 10/n
LIVI can detect trans effects missed by conventional single-gene tests. Such effects act at the level of gene regulatory networks and/or manifest in cell type subpopulations. Example: rs12550612 impacts JAK-STAT signalling in a subpopulation of naive CD4+ T cells (referred to as "LIVI subset"). 9/n
Applied to a dataset of 1M+ PBMCs from ~1000 donors (OneK1K), LIVI recovered previously reported (in eQTLGen) trans-eQTL variants. LIVI also outperformed other methods that infer donor-informed representations from scRNA-seq data, but do not explicitly separate donor from cell-state variation. 7/n
Once LIVI is trained, latent donor factors can be tested for genetic associations. To avoid circularity, genotypes aren't used during training. Discovered effects are projected back to single cells via the donor-cell-state interaction model (DxC). Decoder weights indicate implicated gene sets. 6/n
LIVI builds on VAEs to efficiently model gene expression counts from millions of cells and thousands of donors. Unlike standard VAEs, we use a structured decoder architecture to separate variation attributable to cell-states vs. donor characteristics. 5/n
To address this limitation, together with Tobias Heinen, @manusaraswat.bsky.social , @brianfclarke.bsky.social and @oliverstegle.bsky.social , we devised LIVI. LIVI enables discovery of trans genetic effects at the level of gene regulatory networks and at single-cell resolution. 4/n
Despite their disease relevance, trans-eQTLs remain understudied compared to cis effects. This is largely due to limitations of conventional methods, which test single genes, typically pseudobulked, thereby missing complex cell-state-specific trans effects across gene networks. 3/n
Single-cell eQTL studies can reveal how disease-associated variants disrupt gene regulation in individual cells, enabling detection of effects in rare, disease-initiating cell populations that ultimately give rise to disease phenotypes (illustrated beautifully in www.nature.com/articles/s41...). 2/n
Delighted to present Latent Interaction Variational Inference (LIVI), a framework for trans-eQTL mapping at single-cell resolution that I developed during my PhD together with colleagues from @steglelab.bsky.social 1/n
🧬 New preprint alert! After years of collaborative work across 52 datasets we are presenting eQTLGen phase 2: a genome-wide eQTL meta-analysis covering 43,301 blood samples: www.medrxiv.org/content/10.6... (1/8)
Fresh off the press in 2026! Interested in the challenge of how to advance from descriptive atlases to causal mechanisms and counterfactuals? 🔬
Take a look at our recent perspective: "Interpretation, extrapolation and perturbation of single cells"! (rdcu.be/eXeDY)
Back from #scverse25 and feeling inspired by what @scverse.bsky.social has achieved over the last years. Impressive tools, a vibrant community and we even had a reunion with our two alumni Danila and Max Frank.
We also presented our latest findings from the lab 👇
🧠 Excited to share my main PhD project! We mapped the regulatory rules governing Glioblastoma plasticity using single-cell multi-omics and deep learning. This work is part of a two-paper series with @bayraktarlab.bsky.social @oliverstegle.bsky.social and @moritzmall.bsky.social, Preprint at end🧵👇
Segger logo
Message passing intuition behind the segger’s link-prediction model and the network architecture
1/ New preprint! 🍳
@elihei.bsky.social and our team at @embl.org , @dkfz.bsky.social, and @mskcancercenter.bsky.social built #segger - a fast, accurate cell segmentation tool for spatial transcriptomics that assigns transcripts to their cell origins!
doi.org/10.1101/2025...
PHD POSITIONS IN COMPUTATIONAL RESEARCH, DATA SCIENCE AND ARTIFICIAL INTELLIGENCE Are you interested in solving fascinating biological or medical questions computationally? At the German Cancer Research Center (DKFZ), Germany’s largest biomedical research institute with its core site in Heidelberg, computational scientists work at the forefront of artificial intelligence and cancer research. They combine inter disciplinary approaches from computer science, medical informatics, physics, biology, bioinformatics and statistics to analyze and understand complex biological and medical data. Main activities at the DKFZ involving big data: • Development and application of machine learning and computa tional methods for multi-modal data integration, including integrative analysis of genomic, epigenomic, transcriptomic, proteomic, meta bolomic and radiomic imaging data • Development of methods for modeling and simulation of biological and medical processes • Medical Informatics • • • • DNA mutation analysis and machine learning to decipher cancer genome variation Application and development of state-of-the-art technologies for single-cell omics and life cell imaging Computer-assisted radiology and surgery Biostatistical evaluation of experimental and clinical data More information at www.dkfz.de/datascience. Full funding is provided for the duration of the PhD. To apply online to the International PhD Program visit To apply online to the International PhD Program visit www.dkfz.de/phd Application deadline 24 April 2025 We support and value diversity and encourage applications from individuals from underrepresented groups. If you require assistance during the application process, please contact phd@dkfz.de.
The next selection round for the DKFZ International PhD Program has started. Please consider applying to this really excellent program! I'll be proposing projects, as will many other outstanding PIs.
