Atom-JEPA-Bio
Model contribution
I optimized and ran Atom-JEPA-Bio’s second-stage training on LUMI’s AMD GPUs. We completed ten epochs in about ten hours on four nodes.
Machine learning for molecular and biological research.
I develop, train, and evaluate machine-learning models, with recent work on molecular representation learning and bioassay-supervised pretraining.
I am a postdoctoral researcher at DTU Compute through the end of 2026. My PhD thesis is submitted, and I am awaiting my defense.
I am interested in ML scientist and research engineer roles in Copenhagen.
Molecular models & generation
-
-
Semi-Supervised Learning for Molecular Graphs via Ensemble Consensus
International Conference on Machine Learning (ICML).
Changing a molecule can change its properties. We use agreement on unlabelled molecules to learn without augmentations, with individual models often outperforming conventional ensembles.
-
Same Graph, Different Likelihoods: Calibration of Autoregressive Graph Generators via Permutation-Equivalent Encodings
AISTATS Workshop: Towards Trustworthy Predictions — Theory and Applications of Calibration for Modern AI.
Using AutoGraph’s SENT encodings ↗, transformers learn to generate molecular graphs as strings. On QM9, we found that comparing likelihoods across equivalent strings provides a better quality signal than generation likelihood alone.
Evaluation & uncertainty
-
Calibrated Uncertainty Quantification on TDC ADMET: What Can Be Resolved Across 22 Datasets?
AIDaR Workshop at NeurIPS. Camera-ready version forthcoming.
Across 22 ADMET datasets, we compare eleven calibrated prediction pipelines for accuracy and uncertainty estimates. We test whether the apparent differences between methods hold up across chemical splits, model seeds, and smaller evaluations.
-
On Joint Regularization and Calibration in Deep Ensembles
Transactions on Machine Learning Research (TMLR).
Are we training deep ensembles the right way? Stopping based on joint performance let members train longer and improved ensemble accuracy and predictive likelihood in our experiments.
Training & optimization
-
ICICLE: In-Context Inference for Curve-Learning Extrapolation
AXIOM: Foundations of Efficient Deep Learning — NeurIPS Workshop. Camera-ready version forthcoming.
Building on LC-PFN ↗ and FT-PFN ↗, I developed a model that predicts learning curves to guide which training runs to pause or resume. Its encoder handles varying numbers of hyperparameters without fixed-length padding. ICICLE integrates with Ray Tune.
-
torch-distributed-ensemble
Research software
I designed a training setup that keeps ensemble members independent across GPUs while making their joint performance easy to track.
Experimental biology
-
Head-to-head comparison of modular vaccines developed using different capsid virus-like particle backbones and antigen conjugation systems
Vaccines.
I combined vaccine experiments with quantitative analysis of IgG responses, comparing AP205 and HPV particle backbones and covalent versus affinity-based antigen attachment. This work formed the basis of my MSc thesis.
-
Capsid-like particles decorated with the SARS-CoV-2 receptor-binding domain elicit strong virus neutralization activity
Nature Communications.
As part of the vaccine development work, I cloned, expressed, and purified candidate Spike and receptor-binding domain (RBD) antigens using a baculovirus–insect cell system.
Scientific applications
At LEO Pharma, I extended an existing Django/React application for omics data and built RNA-seq analysis pipelines, working with domain experts on the tools they needed.
For DaSSCo’s digitisation of natural history specimens ↗, I created a Python package for printable Data Matrix labels ↗ for pinned insects and extended existing camera-control and imaging software.
I scaled multi-node training of NequIP and Allegro interatomic potentials on LUMI. I also built boltz-lumi ↗, which adds ROCm kernels for Boltz-2 inference on AMD GPUs.
Education & research
MSc and BSc in Molecular Biomedicine, and BSc in Machine Learning & Data Science, University of Copenhagen.
My experimental experience includes vaccine antigen cloning, expression, and purification.
My submitted PhD thesis is titled “Uncertainty and Evaluation in Molecular Machine Learning”. I am awaiting my defense.
Teaching
Teaching assistant at DTU:
- MLOps ↗ January 2024 + 2025 Reproducibility · Deployment · Model maintenance Course website ↗
- Deep Learning ↗ Autumn 2024 Neural networks · Transformers · PyTorch
- Advanced Machine Learning ↗ Spring 2024 Generative models · Geometric representations · Graph neural networks