As a structural Bioinformatician, I spend most of my time working with proteins that exist only as lines of code and clouds of atoms on a screen. My work focuses on computational structural analysis through modelling, docking and simulation, to understand the behaviour of protein in real world. It sounds abstract, and in some ways it is, but for me, every structure is a puzzle with real implications, especially when we’re designing antibodies that might one day help patients.
Before AI entered the picture, this work was slow, steady, and often frustrating. Building proteins atom by atom, loop by loop, required endless patience. Simulations could run for days just to reveal that a design was unstable. A single mutation in a sequence could collapse everything. Much of my time was spent wrestling with failures, learning more from what didn’t work than from what did.
"With AI, instead of starting from scratch, I had powerful tools that could suggest structures already close to reality.
That changed when AI started to shape our field. Suddenly, instead of starting from scratch, I had powerful tools that could suggest structures already close to reality. AlphaFold taught machines to “read” the language of proteins. RFdiffusion opened possibilities to generate entirely new backbones.These weren’t shortcuts; they were new starting points. They allowed me to focus less on brute-force trial and error and more on asking meaningful questions about stability, binding, and therapeutic potential.
I have seen this shift firsthand in my antibody design work. Where I once struggled to generate a handful of backbones worth simulating, AI now provides me with a set of designs that look promising from the outset. That doesn’t mean the work is finished; far from it. I still have to run stability checks, perform binding free energy calculations, and refine candidates through molecular dynamics. And in the end, the lab decides whether any of these digital molecules hold up in reality. But the difference is profound: the distance between imagination and testable design has narrowed.
What AI has really changed is the kind of questions we dare to ask. Can we design antibodies completely from scratch? Can we predict failure before a single wet-lab experiment is done? Can we tune therapies so precisely that they feel tailor-made for each patient? A few years ago, these were dreams. Today, they’re daily conversations.
"And yet, some things haven’t changed. There are still late nights when I stare at a protein model waiting for it to fold, wondering if it will ever behave. There are still moments when everything collapses in simulation. But there are also quiet victories when an antibody binds tighter, when a structure holds steady, when a design whispers yes instead of no.
And yet, some things haven’t changed. There are still late nights when I stare at a protein model waiting for it to fold, wondering if it will ever behave. There are still moments when everything collapses in simulation. But there are also quiet victories when an antibody binds tighter, when a structure holds steady, when a design whispers yes instead of no.
That’s what keeps me here, at the boundary of atoms and algorithms. AI hasn’t replaced the heart of structural bioinformatics; it’s amplified it. It gives me more chances to listen to proteins, to learn from them, and to turn those lessons into something meaningful.
- Bioinformatics
.png)
.jpg)