By 1:30pm on Friday, May 15, I had been talking for over an hour straight. Not in a lecture hall, not on a podcast, but standing in front of a research poster at the University of Washington’s 2026 Undergraduate Research Symposium, explaining almost half a year’s worth of work to anyone who stopped long enough to listen. My jaw hurt by the time it was over.

 

It was absolutely worth it.

 

The Undergraduate Research Symposium is one of UW’s most celebrated events of the year. Hundreds of student researchers presenting across every field you can imagine at a university, with hundreds more walking through as guests. I was both. Of course, I spent time at my own poster and then wandered around the room after presenting, stopping at other people’s work, asking questions, trying to understand what a geology student or a public health researcher or a computer science undergrad had spent their year figuring out. Fascinating things to see from researchers early in their careers.

 

That part alone was worth showing up for.

 

But the real thing, the part I keep thinking about, was what happened at my own poster.

​

 

So what have I been working on as part of my research project?

 

Over the past academic year in the Elbert Lab, I built a quantitative systems pharmacology model of amyloid-β dynamics in Alzheimer’s disease, stratified by the APOE4 genotype. APOE4 is the most common genetic risk factor for late-onset Alzheimer’s, and its primary mechanism is through microglia— the brain’s resident immune cells—impairing their ability to clear amyloid-β plaques. The goal of the model was to generate virtual patients: computational representations of APOE4 carriers and noncarriers whose simulated biomarker trajectories were calibrated against real clinical brain scan data (through PET scans) and cerebrospinal fluid measurements.

 

In plain English, we built stand-ins for real patients out of math and code, trained them on real data, and used them to study how a single gene changes the way Alzheimer’s unfolds in the brain and responds to anti-amyloid therapies.

 

The question I kept getting asked at the poster (one I certainly most enjoyed answering) was why. Why build virtual patients at all?

 

Alzheimer’s disease begins a decade or more before the first symptom surfaces. The amyloid plaques that eventually destroy memory and cognition start accumulating silently, long before anyone notices anything is wrong. By the time a patient is diagnosed, the disease has been running for years. Running clinical trials long enough to capture that full arc of progression isn’t just expensive. At the scale we actually need, it’s almost logistically impossible. You’d be asking patients and researchers to commit to decades of follow-up, and even then you’d only capture one trajectory per patient.

 

Definitely not sustainable.

​

​

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

​​

​

Amyloid-β plaques (blue circular structures) entangled around a neuron inside the brain.

 

Virtual patients, however, change that equation. Once you’ve validated a computational model against real clinical data—which is what calibrating against OASIS-3 amyloid PET scans and LOAD cerebrospinal fluid measurements was all about—you can simulate disease progression across different patient subgroups without waiting decades for the data. You can test how APOE4 carrier status modifies the accumulation rate of amyloid-β. You can model how a specific therapeutic might work differently in carriers versus noncarriers. You can generate hypotheses that would otherwise take thirty years to answer, and you can do it before committing to a full clinical trial.

 

This is not a shortcut. It is how you get ahead of a disease that moves gradually and in silence. Predictive disease modelling is part of how Alzheimer’s research eventually catches up to Alzheimer’s itself.

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​​​

The Open Access Series of Imaging Studies (OASIS) is a project aimed at making neuroimaging data sets of the brain freely available to the scientific community. | Credit: OASIS by Washington University in St. Louis

​​​​

​

What I didn’t expect going into Friday was how much I’d learn from explaining it all out loud, over and over, to people with completely different levels of familiarity with the field. You think you understand your own work until someone who has never heard of spline regression asks you to explain it from scratch, or until someone who clearly knows the field pushes back on your methodological choices and you have to defend them in real time. That was the real test and the most valuable learning experience.

 

Explaining why I only adjusted microglial clearance parameters between APOE4 carriers and noncarriers, rather than also adjusting aggregation and production, is a different kind of intellectual exercise when you’re standing at a poster and someone is scrutinizing you. You either know your reasoning or you don’t. You either understand your own limitations or you don’t. Those conversations taught me more about my own research than any week of running the model did. But, in the end, I had a proper explanation for it by describing the sensitivity analyses I ran, validated by the model represented on the poster itself.

 

I also didn’t expect to find it as energizing as I did. An hour of continuous talking should be exhausting. And physically, it was. My jaws hurt and I didn’t want to talk much after that lol. But there is something about sharing work you truly care about with people who are curious enough to stop and engage with it that doesn’t feel like performance. It feels like the point of it all.

 

 

Here’s what I’ve learned over the past several months that no course has ever taught me as cleanly…Research is not about the results. Or rather, it’s not only about the results. There were times this year when the model wasn’t converging, where the optimization kept failing to reproduce what the clinical data said it should. In those moments, the obvious move feels like abandoning the approach and starting over. The less obvious but more important move is to go back to the assumptions. To interrogate what you believed about the system and figure out where the belief broke down.

 

I ran the model and it didn’t work. I changed something and ran it again. I did that more times than I can count. What I got from that process wasn’t just a better model. I got a habit of mind that I think will outlast this project by a long time. I’ve come to understand how important this is in research especially.

 

The willingness to stay with a question through failure, to treat a poor result as beneficial information rather than defeat…that is the skill. The specific findings are secondary.

 

This research is still very much a work in progress. There are real limitations I acknowledged on the poster, the biggest being that we only modelled APOE4’s effect on microglial clearance, not on aggregation or production, which are also affected. In addition to this, incorporating tau dynamics, which is what ultimately leads to neurodegeneration, is absolutely essential to making this predictive model even more accurate.

 

The next layer of the model is already clear. There are many ways to still improve this. The work continues.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

​My research poster summarizing my work since September running sensitivity analyses and creating APOE4-stratified virtual patients for Alzheimer’s disease

​​​

 

Check out an interactive, zoomable version of the poster above.

 

A huge thank you to Dylan Esguerra and Dr. Don Elbert for being exceptional mentors this year. I have learned more from both of you than I can adequately summarize here, and I’m very excited to keep going as part of the Elbert Lab to work towards advancing Alzheimer’s research.To everyone who stopped by at my poster on Friday, asked me difficult questions, and pushed me to think more carefully about my own work: that is exactly what a research symposium is for, and you made it worth the sore jaw.

 

Already looking forward to the next one. 🧠