Liaison between SEVO Bioscience and Mila

SEVO Bioscience hired me as the liaison between the company and Mila, the Quebec AI Institute, to build an AI algorithm for protein design. I also consulted on the computational biology side.

The purpose

The goal was to use AI to design better proteins for downstream applications. Instead of testing variants one by one in the lab, an AI-driven workflow can propose new designs, predict their structure and estimate how well they bind their target, all before a single experiment. The lab then tests only the most promising candidates.

The reason this matters is scale. A protein of a few hundred amino acids can be changed in an astronomical number of ways. Even a well-equipped lab can only make and test a tiny fraction of those variants. Traditional approaches rely on careful guesses and many rounds of trial and error. Computational design does not remove the need for experiments, but it changes where you start, so that each experiment is far more likely to teach you something useful.

Why AI, why now

Biology has changed quickly. In 2024, the Nobel Prize in Chemistry went to David Baker for designing new proteins, and to Demis Hassabis and John Jumper for predicting protein structures with AI. Problems that took years of lab work can now be explored on a computer in days.

Structure prediction was a turning point. For decades, solving a protein’s shape required difficult experimental methods. Now a reasonable model can often be generated quickly. Combined with generative tools that propose new sequences, that opens the door to designing proteins with specific functions rather than only studying the ones nature already made.

What a biologist brings

AI models are only as useful as the questions they are asked.

AI models are only as useful as the questions they are asked. A biologist knows which questions matter, which results are plausible, and which experiments can test a prediction. My role was to build that bridge: between what a model suggests and what a cell actually does.

That bridge matters because a predicted structure is not a working protein. A protein also has to fold properly when it is produced, stay stable, avoid unwanted immune responses, and do its job inside a complex biological system. Many of those properties are still hard to predict. Knowing where the models are strong and where they are weak is what keeps a project grounded.

Two kinds of thinking

Working with Mila’s researchers, who think in models rather than pipettes, changes how you design experiments. It also forces the AI side to stay honest about what the biology can and cannot confirm.

On the biology side, it pushes you to collect data more systematically, because a model learns from what you measure and how consistently you measure it. On the computational side, it is a reminder that a high score from a model is a hypothesis, not a result.

My perspective

I don’t see AI replacing lab work. I see it changing what lab work is for. Less time spent searching blindly, more time spent testing ideas that are already well reasoned. The scientists who will benefit most are the ones who can move comfortably between both worlds.

Montréal is one of the few cities where world-class AI and life sciences sit a few metro stops apart. Partnerships like this one are how they meet.