Entry 001 · Welcome

Hi, I'm Alan. I'm interested in searching for host-specific patterns imprinted in viral genomes.

I'm a researcher at Wayne State University School of Medicine, working with Dr. Phil Pellett. Most recently, that meant building a 58,046-genome dataset across 15 virus families and training neural networks on raw k-mer frequency (no alignment, no annotation) to tell human-infecting viruses from everything else. This site is where I keep track of what I'm building and learning along the way.

01

A place to keep track of what I'm working on

This is basically my lab notebook, public. I write up experiments and side projects as I go, mostly around AI and virology, partly to keep myself honest and partly because I like sharing this stuff.

A

Building datasets worth training on

GenBank's metadata is a mess: inconsistent host labels, missing fields, free-text notes. I built a three-tier pipeline (string matching, pattern lookup, then an LLM for the ambiguous cases) to turn 82,513 raw genomes into a clean, host-labeled set of 58,046.

B

Reading the model's reasoning

Predicting host compatibility is only half the point. I ablate individual k-mers to see which ones actually move the needle, then map those back onto genes to find out what the model is really picking up on.

ML

Host-range prediction

Neural nets on k-mer frequency vectors (k = 3 to 8) beat logistic regression and random forest baselines at telling human-infecting viruses from the rest.

DNA

Genome generation

Projects like MambaVirus and SOMBRA that generate and simulate viral genomes, rather than just classify existing ones.

EPI

Adaptation over time

Tracked SARS-CoV-2's predicted human-adaptation signal drifting upward across 8,683 genomes from 2019 to 2024.

02

A few things I've been building

All projects

Some of the projects I've worked on, mixing research with a bit of design and storytelling.

03

Notes from the lab

Archive

Things I've learned or thought about while working on AI and biology.

04

Get in touch

Have a question or want to say hello? I'd love to hear from you.