Selected work

An agentic environment for genome-scale metabolic model curation. A set of defined functions and AI skills lets the user drive model validation and curation with minimum code interaction, so the scientist owns the biology while AI runs the code.

It grew out of Master Thesis work: the same validation and curation cells were being re-run by hand with slightly different parameters to work out why a model was not behaving as expected.

The repository is public and the tool is usable today; validation against curated models is still in progress, so its output is best read as leads to check.

A set of bash and R scripts, split across four modules, that run sequentially to cover the full analysis path for WGS, RNAseq and ATACseq - from raw, demultiplexed reads through to multiomics data integration and plotting.

It was built to analyse a specific experiment on drug-resistant cancer cells, where integrating the three data types was essential to read the effect of a treatment. It is now config-driven: running it on new data means editing one config file and the sample sheets.

An ordered single-cell RNA-seq workflow taking a provided dataset from quality control through to annotated cell types using a reference model.

Produced as an assignment for the Genomics and Transcriptomics Analysis lecture. The notebook is complete for the task it was set; the JupyterLab environment it ran on was provided by the course and is not part of the repository.

Six classifier families, benchmarked against a majority-class baseline, predicting whether a protein is detectable by LC-MS/MS from five sequence-derived and transcript-level properties, on the A. thaliana PeptideAtlas dataset.

Finished and published: the dataset and a requirements.txt are in the repository alongside the notebook.