Our drug discovery paper is going to ACS Spring

Our paper on using topological data analysis for neurodegenerative drug discovery was accepted to ACS Spring 2026, the American Chemical Society’s spring meeting in Georgia. It’s in the Division of Computers in Chemistry, which is a very good name for a division and also a pretty accurate description of what I do.

This is the second time the work has been accepted somewhere. We presented an earlier version at the Southern California Conference for Undergraduate Research in 2025. ACS is a much bigger room.

What the paper is about

Diseases like Alzheimer’s and Parkinson’s are different diseases, but the drugs used to treat them can have a lot in common. We wanted to see that structure directly, across many drugs and many diseases at once.

So we profiled 345 FDA-approved and clinical-stage compounds across seven neurodegenerative diseases. For each one, we built features from 94 physicochemical descriptors and an 881-bit PubChem fingerprint, which is basically a long checklist of which chemical substructures a molecule has. All of it came from five public databases.

That gives you a lot of dimensions, and you can’t look at 975 dimensions. This is where topological data analysis comes in. Instead of asking which drugs are closest to each other, it asks what shape the whole dataset has: which groups hang together, what connects them, and where the gaps are.

We used two tools for that. Mapper turns the data into a graph you can actually look at, and we ran six versions of it with different UMAP and t-SNE settings to make sure the shape wasn’t just an artifact of one setting. Persistent homology measures which features of the shape are real and which are noise, and we ran it separately for each disease.

What we found

The biggest result was that a lot of the structure across diseases comes from metabolism. In four of the seven diseases, interactions with CYP450 enzymes, which break down a large share of drugs in the body, were driving how drugs grouped across diseases.

We also turned the analysis into a reusable chemical space. You can project a new compound into it and see where it lands relative to existing drugs. We validated that with leave-one-out testing, where you remove one compound, rebuild everything, and check whether it lands where it should.

The paper is under review at a journal now. In the meantime, I get to show it to a room full of chemists, which is both exciting and a little terrifying.