Niklas Abraham

I am an MSc student in Machine Learning at the University of Tübingen (current grade average: 1.0), where my studies are embedded within one of Europe’s leading ML research hubs through links with the Max Planck Institutes and Cyber Valley. My coursework focuses on deep learning, statistical machine learning, and probabilistic inference. I completed my BSc in Simulation Technology at the University of Stuttgart in 2025 (grade 1.9, 5th in cohort of 22), with an elected focus on AI and bioinformatics. I am a scholarship holder of the German Scholarship Foundation (Studienstiftung des Deutschen Volkes).

My research interests center on advanced generative modelling (score-based diffusion, flow-matching), non-convex optimisation in deep learning, and reinforcement learning for sequential decision-making, with applications in structural biology and bioinformatics. Most recently, our work on converting diffusion-based co-folding models to deterministic probability-flow ODEs was accepted as a poster at the ICLR 2026 Workshop on Learning Meaningful Representations of Life, achieving a 20x sampling speed-up on FoldBench.

Previously, I worked as a research student at the Institute of Biochemistry and Technical Biochemistry (University of Stuttgart) under the supervision of Prof. Jürgen Pleiss, where I developed machine-learning frameworks integrating large protein language models (ESM-2, ESM-C) with curated biological databases to predict β-lactamase resistance phenotypes, and contributed to the Python Enzyme Engineering Database (PyEED). In October 2025, I won the Speed-Up Challenge at the Merck M-Boltz Hackathon by developing a fast deterministic flow-matching framework for accelerated protein structure generation. Starting April 2026, I will join Merck KGaA’s Group Science & Technology Office for a six-month internship.

Outside my academic work, I train for triathlon (next up: a 70.3 Ironman), coach climbing for children at KiSS e.V. twice a week, and read as much non-fiction as I can find.

News

Apr 1, 2026	Starting my six-month internship at Merck KGaA’s Group Science & Technology Office in Darmstadt. I will be working on metagenomic sequencing pipelines and benchmarking protein language models and DNA transformers for variant effect prediction.
Feb 1, 2026	Our paper “Converting diffusions to flows accelerates sampling and suggests over-conditioning of co-folding models on sequence” has been accepted as a poster at the ICLR 2026 Workshop on Learning Meaningful Representations of Life (LMRL).
Oct 15, 2025	Won the Speed-Up Challenge at the Merck M-Boltz Hackathon in Darmstadt by developing “Boltz Flow Matching”, a fast deterministic flow-matching framework enabling analytical conversion from score-based diffusion to flow equations for accelerated protein structure generation.
Oct 1, 2025	Started my MSc in Machine Learning at the University of Tübingen. Looking forward to diving deeper into generative modelling, probabilistic inference, and reinforcement learning within the Cyber Valley research ecosystem.
Sep 15, 2025	Completed my BSc in Simulation Technology at the University of Stuttgart (grade 1.9, 5th in cohort of 22). My thesis on predicting β-lactamase resistance phenotypes from sequence using protein language models is now listed on the Research page.

Selected Publications

Converting diffusions to flows accelerates sampling and suggests over-conditioning of co-folding models on sequence

Nele P. Quast, Niklas Abraham, Aaron Schöne, Fergus Imrie, Matthew I. J. Raybould, Yee Whye Teh, and Charlotte Deane

In ICLR 2026 Workshop on Learning Meaningful Representations of Life (LMRL) 2026

Abs

Deep generative models can predict protein structures from sequence with high accuracy; however, sampling from these models remains computationally burdensome, with current protocols using hundreds of iterations through the trained model to obtain a final predicted structure. To accelerate sampling and improve the interpretability of the prediction trajectories, we convert the stochastic diffusion sampling process into a deterministic flow process. We show that the conversion of pre-trained, diffusion-based structure prediction models to probability-flow ODEs yields equivalent performance on the FoldBench benchmark alongside a 20x sampling speed-up. Furthermore, we demonstrate the effects on prediction diversity and use the intermediate predictions made along the de-noising trajectory to show that deep generative structure prediction methods are strongly conditioned on the sequence and MSA embeddings, appearing to make predictions with weak sensitivity to the noise initialisation. Finally, we discuss the implications of strong sequence conditioning for generative protein structure prediction and protein design, as well as pointing to future experiments that build on our initial results.
Geometric encoding of enzymatic mechanism in protein language model representations

Jan Range, Niklas Abraham, and Jürgen Pleiss

Manuscript in preparation 2026

Abs

Understanding how protein language models encode biological function remains a fundamental challenge despite their remarkable empirical success. Sequence-level inference typically reduces per-residue embeddings to a single vector via mean pooling, disregarding the internal geometry of the token cloud. We propose that functionally critical residues are geometric outliers within this cloud: their embeddings deviate from the within-protein mean in direction rather than magnitude, reflecting the superposition of mechanism-specific features learned during masked-token pretraining against a structurally generic majority. Here we show that cosine dissimilarity from the within-protein mean identifies these residues without supervision, and that the subspace spanned by their embeddings carries precise functional information. Across the SwissProt proteome this subspace is enriched for active-site and binding annotations and localises to ligand-binding sites in experimentally resolved structures. Its principal direction encodes catalytic mechanism at sub-subclass resolution on sequence-redundancy-filtered enzymes and rotates in a fitness-predictable manner under single amino acid substitutions, without reliance on evolutionary alignments, structural data, or likelihood scoring. These results reveal that protein language models implicitly organise functional information as a geometric minority, separable from the structural majority by direction alone.
Predicting β-lactamase resistance phenotypes from sequence using protein language models

Niklas Abraham

Bachelor’s Thesis, University of Stuttgart 2025

Abs PDF

Protein language models (PLMs) have emerged as powerful tools for predicting protein structures from sequence. However, their ability to predict resistance phenotypes of β-lactamase enzymes remains unexplored. We propose a novel approach to predict β-lactamase resistance phenotypes from sequence using protein language models. We train a protein language model on a dataset of β-lactamase enzymes and their resistance phenotypes, and evaluate its performance on a dataset of β-lactamase enzymes and their resistance phenotypes. We find that the protein language model is able to predict β-lactamase resistance phenotypes with high accuracy, and that the model is able to predict resistance phenotypes that are not present in the training data. We also find that the model is able to predict resistance phenotypes that are present in the training data, but with lower accuracy. We conclude that protein language models are a powerful tool for predicting β-lactamase resistance phenotypes from sequence.