About
When the expertise of a computer scientist and the ambition of a researcher meet the challenges of unraveling large multi-omics datasets, interesting things can happen. In my case, it results in theoretically sound yet pragmatic machine learning methods on tabular and network data. Among other applications, these methods reliably infer interpretable clusterings and trajectories, bridging the gap between classical machine learning and fundamental research.
Explore my research offering practical insights into complex systems in the fields of neuroscience, single cell genomics and spatial transcriptomics.
Degree:
Computer Science M.Sc.
Status:
Researcher at RWTH Aachen, Germany
Keywords:
Machine Learning
Data Science
Optimal Transport
Single-cell Genomics
Metric Learning
Graph Learning
Brain Connectivity
Languages:
C
C#
Python
JavaScript
TypeScript
HTML
MATLAB
CSS
Shell
Projects
Python Package
Supervised Optimal Transport
ggml-ot is a Python package for Supervised Optimal Transport (OT) that improves downstream applications, such as classification, clustering, trajectory inference and embeddings. It is fully integrated into the scVerse Ecosystem streamlining OT-based analysis of single-cell data in the Anndata format. It also provides interfaces to support generic array-like data.
Languages
- Python 100%
Paper
Global Ground Metric Learning
Optimal transport (OT) is a powerful mathematical framework for comparing probability distributions with promising applications in patient-level scRNA analysis. Crucially, the effectiveness of OT is significantly influenced by the choice of the underlying ground metric (or cost). However, predefined metrics typically cannot account for the inherent structure and varying significance of different features in the data. Existing supervised ground metric learning methods often fail to generalize across multiple classes or are limited to distributions with shared supports, which is not the case for scRNA data. To address this issue, we introduce a novel approach for learning metrics for arbitrary distributions over a shared metric space. Our method provides a distance between individual points (cells) like a global metric, but requires only patient-level class labels (e.g., disease states) for training. The resulting learned global ground metric enables more accurate OT distances, which significantly improves clustering and classification performance in synthetic and real-world scRNA datasets spanning various diseases.
Keywords
Optimal Transport
Metric Learning
Single-cell Genomics
Paper
Wasserstein Graph Distance
In this paper we described an unsupervised, optimal transport based approach to define a distance between graphs. The idea is to derive representations of graphs as Gaussian mixture models, fitted to distributions of sampled node embeddings over the same space. The Wasserstein distance between these distributions then yields an interpretable and easily computable distance measure, which can further be tailored for the comparison at hand by choosing appropriate embeddings.
Keywords
Optimal Transport
Graph Learning
Brain Connectivity
Snakemake WMS
Neuroimaging Analysis Pipeline
WIPAR stands for Widefield Imaging Pipeline for Analysis and Regression. It is a data pipeline for processing and analysing task-specific (widefield) calcium imaging data through neural decoding. Here, calcium activity is a proxy for neuronal activations. It provides stand-alone functionalities to visualize the data analysis as well as enabling the export of processed data for other visualization purposes.
Languages
- Python 85%
- MATLAB 10%
- Shell 5%
Non-profit
Nisaba App
The continuous assessment of the impact on various stakeholders is an integral part of development cooperation projects. Nisaba is engaged in the digitization and simplification of these processes, developing a freely configurable software solution for this purpose. This solution is already being used within the projects Aktion Sodis and with initial pilot partners, and will be scaled further as soon as possible.
Languages
- TypeScript 45%
- JavaScript 30%
- HTML 15%
- CSS 10%
Research Focus Class
Eye-Tracking for Learning Analytics
Integration of eye-tracking glasses into the multitouch learning game framework for learning analytics. Our implementation is based on a master-client architecture, where multiple eyetrackers automatically connect to the learning game framework. Eye-tracking data is processed in real-time and mapped to GUI elements. It can be used within the learning games for adaptive feedback or subsequent learning analytics.
Languages
- Python 60%
- JavaScript 40%
Contact
Let's get in contact!
If you want to collaborate or work together, drop me a message. Looking forward to hear from you!