Computational biologist working at the intersection of drug discovery and
large-scale biomedical data, think CRISPR screens, genetics, transcriptomics,
multi-omics. AI has become a bit of a buzzword, but I do enjoy the application
of ML/AI where (from my perception) useful: vector searches, quantisations,
graph-based methods are honestly just cool. Experimenting with Burn as a Deep Learning framework that can run on most GPUs.
Lately spending more time building open-source tooling: R packages for
bioinformatics & computational biology and Rust crates, often combining the two
for performance-critical work. My background is in industry with experience in
larger pharma companies and small biotechs, but overall just a firm believer
that good science needs performant and accessible open-source software that does
not necessitate fat cloud servers. Some Python packages also in the making
around ETLs of large-scale biomedical data. Keep an eye open...
Below are R packages I created and maintain on the side as labour of passion:
| Package | Description |
|---|---|
| bixverse | The tidyverse equivalent for bioinformatics... Highly accelerated and optimised computational biology methods. Also contains a single cell analysis framework enabling million cell analyses on small compute. |
| bixverse.gpu | SIMD-accelerated CPU code not fast enough? Need more oomph? Contains GPU-accelerated methods (mostly for single cell) for the bixverse (and also manifoldsR in form of a parametric UMAP). |
| bixverse.plots | Plotting helpers for bixverse because codebase is already too large... Very early days for that package. |
| genewalkR | All types of graph-based computational methods; has also an implementation of node2vec and methods based on that small neural network. |
| manifoldsR | 2D embedding and manifold learning method (think single cell visualisations) in a single package made blazingly fast via Rust. |
Close to dying my hair blue... Jokes aside, below are the public Rust crates I have created and maintain. These power the libraries and packages in the interpreted, dynamically typed languages, see above. There are here as standalone libraries, so, if you wish to integrate them into your Rust code, all MIT licensed.
| Crate | crates.io | Description |
|---|---|---|
| node2vec-rs | First trials in Burn... Implements node2vec in Burn, but also highly optimised specialised CPU version written in there. | |
| ann-search-rs | Various computational biology application need fast (approximate) nearest neighbour searches and I fell in love with that field/methods... Has highly optimised CPU indices, quantised versions and GPU-accelerated ones. | |
| manifolds-rs | Implementations of the 2D embedding learning methods that power the manifoldsR package. | |
| bixverse-rs | The Rust code powering the bixverse... Initially part of the bixverse package itself, now abstracted out in its independent and growing crate. Future plans include exposing parts to Python. | |
| evoc-rs | Rust port of the EVoC clustering clustering algorithm from the brilliant Leland McInnes. |