Al Depope, PhD
Postdoctoral Researcher · Statistical Genetics & Machine Learning
Institute of Science and Technology Austria (ISTA) · Klosterneuburg, Austria
I build mathematically grounded machine learning for genomics — Bayesian models and inference procedures that scale to biobank-sized data, and the high-performance code that makes them run.
About me
I am a Postdoctoral Researcher at the Institute of Science and Technology Austria (ISTA), co-affiliated with the Medical Genomics group led by Prof. Matthew Robinson and the Data Science, Machine Learning, and Information Theory group led by Prof. Marco Mondelli. I completed my PhD at ISTA in July 2026, in the same two groups.
My work sits at the intersection of mathematically grounded machine learning, numerical mathematics, software development and genetics. During my doctorate I developed gVAMP, a family of approximate message passing algorithms for Bayesian inference in very high dimensions, and used it to run the largest joint genome-wide association study to date — 17 million whole-genome sequence variants analysed jointly, reaching 46% out-of-sample prediction accuracy for human height. The same machinery now carries over to methylation data, proteomic time-to-event models and cross-ancestry polygenic risk scores.
Before ISTA I obtained a BSc in Mathematics and an MSc in Mathematical Statistics from the Department of Mathematics, University of Zagreb.
- Bayesian inference
- Approximate message passing
- High-dimensional statistics
- Polygenic risk scores
- Survival analysis
- High-performance computing
- C++ / CUDA / MPI
- Python & PyTorch
- 17MWGS variants jointly modelled
- 46%prediction accuracy for human height
- 10×speed-up over MCMC baselines
- 40+talks and presentations given
What I am working on
All projects-
TLgVAMP — transfer learning across ancestries
Polygenic risk scores transfer poorly to populations that are under-represented in genomic studies. TLgVAMP borrows strength from large, well-powered cohorts while keeping population-specific signal separate.
-
gVAMP — joint modelling of whole-genome sequence data
An approximate message passing paradigm that fine-maps WGS variants and gene burden scores conditional on all other measured variation genome-wide. Published in Cell Genomics.
-
vampW — proteomic survival analysis
A scalable Bayesian framework for disease onset times. On the UK Biobank Pharma Proteomics dataset it improves onset prediction by 26–33% relative to penalised Cox and deep-learning baselines.
Latest updates
All publications-
Our second paper with the TU Wien Textile Recycling Group, on deep eutectic solvent formulations for recycling elastane-containing textile waste, appears in Waste Management. [DOI]
-
I completed my PhD at ISTA and started as a Postdoctoral Researcher in the same institute, continuing to develop TLgVAMP, a transfer-learning framework for cross-ancestry polygenic risk scores.
-
‘Joint modeling of whole-genome sequencing data for human height via approximate message passing’ is published in Cell Genomics 6(5):101162. [DOI]
-
Our preprint on joint variable selection for omic biomarkers in time-to-event data is posted on bioRxiv. [Preprint]
-
‘Joint Variable Selection in Proteomics Survival Models’ is accepted at the ICLR Workshop on Machine Learning for Genomics Explorations, joint first-authored with Jakub Bajzik. [OpenReview]
-
‘Deep eutectic solvent as a solution for polyester/cotton textile recycling’ appears in Waste Management, an output of a cross-field statistics collaboration with TU Wien. [DOI]
-
Started a machine learning and privacy internship at the University of Vienna, building a modular library for benchmarking membership inference attacks on large language models.
-
Our work extending the gVAMP framework to whole-genome sequence analysis of human height was accepted for an oral presentation at the 7th International Conference of Quantitative Genetics. Our work on the summary-statistics version of gVAMP was accepted as a poster, presented by Jakub Bajzik.
-
I presented ‘Inference of Genetic Effects via Approximate Message Passing’ at ICASSP 2024. [Slides]
-
AwardBest student presentation at the 52nd European Mathematical Genetics Meeting in Vienna. [Slides]
-
The second version of the preprint ‘Light-speed whole genome association testing and prediction via Approximate Message Passing’ is posted on bioRxiv.
Beyond research
I hike whenever the calendar allows — my most recent peak above 4,000 metres was Mauna Kea in Hawaii — and I run. When there is time, I volunteer by preparing and giving competitive mathematics lectures to gifted students at my old high school and at MNM summer camps.
-
On the hike up to the Triftbrücke, Switzerland.
-
On the way to the Gspaltenhornhütte, Switzerland.
-
On the summit of the Schesaplana, above the Lünersee.
-
Below Säntis in the Alpstein, Switzerland.