Tim O’Donnell is a member of technical staff at Open Athena’s Boston office, where he works on structural biology models like MarinFold. In this Q&A, he gets into the reason AlphaFold was an AI turning point before the release of ChatGPT, explains how Open Athena acts as a “third place” between industry and academia, and shares his favorite hobby.
What’s one of your earliest experiences with AI technology?
Tim O’Donnell: There’s machine learning, and then there’s the big models with really mind-expanding outcomes that we have right now. So for machine learning, my career has always been at the intersection of biology and machine learning. But a big moment for me, for sure, was AlphaFold winning the Critical Assessment of Structure Prediction (CASP) in 2020. I had worked in structural biology toward the beginning of my career, and it really felt like a far-off day that we’d be able to predict protein structure computationally. So it was very memorable for me, seeing those results, and then eventually reading the paper. And I think everyone reading that paper, we all just started imagining all the things you can build with the toolbox that DeepMind unveiled. So much would open up in terms of being able to computationally design molecules, think about drug/target interactions, and do bioinformatics at a new scale, where we’re now including structure proteome-wide in the picture. Later on, we had the 2022 ChatGPT moment. I now see that this 2020 result was the first hint that this decade is going to be pretty different than the one before.
What drew you to work at Open Athena?
Open Athena, for me, is a third space between academia and industry, where we have the broad, socially beneficial mandate that academia has, but we have resources that are very hard to acquire in an academic lab. I have spent a bunch of time in academic groups, and I realized a couple years ago that the scale of compute that’s required to train the models that we’re all using is just really inaccessible, except for maybe the very biggest labs. So Open Athena’s a chance to get to work on the core research processes behind these models—training them and evaluating them—but without being behind the firewall that industry erects. We’re able to share our work widely.
What Open Athena project do you primarily work on?
Right now, I’m mostly working on a new project called MarinFold. The idea is to see, for AlphaFold and related models for structural biology, how much of that can we distill into a vanilla LLM architecture? We have a few reasons to try this: One is that this may open up new avenues for inference time search and for post-training, using ideas that are now fairly standard for LLMs. And then another longer-term idea is that this may give us hints on how to integrate molecular data with language models trained on natural language text, including the molecular structure as a modality. There might be advantages in allowing the agent you’re working with to see the structures you’re thinking about.
Did you have any experience in structural biology before you began working on the project?
My background is more in immunology and cancer. So my PhD is in immunology, where I worked on very simple machine learning models for predicting what would be a good thing to put into a vaccine— what would likely get a T-cell response. And we used that mostly for personalized cancer vaccination, but also for some infectious disease work, especially around COVID. Structural biology is still a new field for me. I worked in it briefly, in the beginning of my career, doing molecular simulation. But I didn’t come back to it until the AI methods really started working.
What sort of non-project specific work do you do at Open Athena?
Part of our goal with projects like MarinFold is to push the Marin project—to help exercise the infrastructure that’s being created for training these language models. So I try to help there. It’s more aspirational right now, but it would be great if projects like MarinFold can make the Marin infrastructure more robust by pushing it in new ways.
What do you like best about working here?
People are always my top priority when figuring out how to spend my time. I think the people here are all kind, generous people who have been pretty thoughtful about where they want to work and what they want to work on. So I’ve really enjoyed the teammates I’ve had here.
What does a typical work day look like for you?
My days right now are a mix. I’m trying to stay engaged in the rapidly changing landscape for molecular design and structure prediction, so there’s a bunch of different forums that I stay a part of to do that. We’re working totally in the open on MarinFold—we haven’t publicized it much, but all of our work is available online day-to-day, and you can see our training runs. We’ve recently begun nucleating a community around that—last week, we created the Discord for it. So another part of my day is coordinating that, talking to folks who are interested in getting involved. And then the majority of my day is running experiments, which is mostly talking to agents now, looking at the results, talking with folks here, and deciding on the next experiment to run.
Do you have any favorite AI tricks or tips, whether for home or for work?
An advantage for me of working at Open Athena is that the folks around me are way more proficient users of all this stuff and generally follow it more closely. So learning from what other folks are doing—that’s how I’m applying AI now. More generally, I’m just trying to stay thoughtful about when I’m using it and what I’m asking it to do, steering it aggressively when it needs to be steered, and trying to trust it the right amount. But the right amount is also changing with every model release.
Do you have a favorite piece of science trivia?
A drop of blood has millions of distinct antibodies. Each of these is the result of an immune reaction you had, to something - virus, allergen, cancer, a new commensal you picked up from your dog, an environmental toxin, autoimmunity - things you are responding to right now and also stuff from your distant past. Long term I'm really excited by the idea that the molecular AI models will let us make sense of that repertoire. There's just so much information on health and disease to uncover if our models get accurate enough to interpret that signal.
Can you tell me about a hobby or activity you enjoy outside of work hours?
Lately, my favorite hobby is taking my 2-year-old daughter on bike rides. I moved to Somerville, Massachusetts last year, and coming from Brooklyn, it’s just amazing what a relaxing, joyful, safe environment it can be to bike around here. I’ve really enjoyed all the opportunities for exploration.
Do you have a signature response emoji, or one you use most often?
Maybe the prayer one. I take it to mean “thank you,” but I also like to imagine that we’re all giving little bows to each other all the time, just in the virtual realm.
Cite this post
@misc{bushwick2026_meet_our_team_tim_odonnell,
author = {Bushwick, Sophie},
title = {Meet our team: Tim O'Donnell},
year = {2026},
month = {oct},
howpublished = {\url{https://www.openathena.ai/blog/meet-our-team-tim-odonnell/}},
note = {Open Athena Blog}
}