Projects
Research pipelines I have built, the computing behind them, and two things you can try on this server right now.
Proteome-wide off-target screening
Research pipeline
Where else does a drug bind? This pipeline screens a set of approved kinase inhibitors against every predicted protein structure in the human proteome and in a nematode model organism.
- Predicted structures for each proteome.
- Pocket detection and pocket-similarity search against the known binding site.
- Inverse docking of each compound into the similar pockets, ranked by ligand efficiency.
- Network scoring to find candidates in central positions of the interaction network.
- Cross-species expression data to judge which candidates are biologically plausible.
The output is a ranked list of candidate off-targets for experimental follow-up. The pipeline is designed so that its predictions can be tested in human cell lines and in the nematode.
Synthetic receptor design
Research pipeline
Molecularly imprinted polymers are synthetic receptors shaped around a target. This pipeline designs candidates for protein targets entirely in silico.
- Enumerate chemically valid oligomers from a monomer set with reaction rules. Chains grow breadth-first or depth-first, and the work is spread over cluster job arrays.
- Dock the whole library against the target with GPU-accelerated docking.
- Profile the interactions of every pose and cluster similar binding modes.
- Prepare the best candidates for molecular dynamics.
Candidates are ranked by sorting on raw, observable metrics one after another instead of a weighted composite score. Every step of the ranking can therefore be checked by hand.
Faster molecular dynamics on GPUs
Benchmarking study
Long simulations are only useful if they finish. This study measured how hardware choices and run settings change simulation throughput on GPUs, so that compute time goes into sampling instead of waiting. It received a best poster award.
Lab computing and homelab
Ongoing
I have built and looked after lab workstations and a small computational lab, from hardware to job scheduling. At home I run a homelab where I test services before I trust them, including self-hosted open-source language models.
A personal site on a small cloud server that I set up by hand, with no hosting platform or page builder. Caddy handles HTTPS and security headers, systemd keeps the services running, and a small collector records the server's health every five minutes.
See how a visit travels or check the server's status.
Small programs that run entirely in your browser. A molecular dynamics sandbox shows atoms condensing and evaporating, a viewer renders molecules in 3D, and there are a few games. Nothing you do there is sent to the server.
Open the playground