Picture two proteins as puzzle pieces. You want to know if they’ll fit together, and what they’ll look like when joined – before you’ve even picked them up. That’s the question I’ve spent most of my days trying to answer at TernaryTx: will these two proteins actually interact?
I’m Holly, a third-year student on the MRC Doctoral Training Partnership in Interdisciplinary Biomedical Research at the University of Warwick, where my research sits between computational work (molecular dynamics simulations) and experimental in vitro wet-lab experiments, looking at how bacterial cells elongate. Before that, I studied Biochemistry at the University of Birmingham. During my PhD, I wanted to gain experience in industry-style research and sharpen my machine learning skills alongside it – so when this internship at TernaryTx came up, it felt like the perfect place to do both.
Most new Machine Learning (ML) models answer that puzzle-piece question by working from sequence information alone – the details encoded in each individual piece, without knowing what the whole structure looks like from the outside. Structure-based models, including co-folding models like AlphaFold 3 and Boltz2, take a different approach: they use the proteins’ overall shape, which lets them compare a new pair to similar-shaped ones that already exist – a bit like recognising two puzzle pieces because you’ve seen ones just like them click together before. Sequence-only models don’t get that shortcut. That’s more hard work upfront, but it also means we’re not thrown off when there’s nothing similar to compare against – particularly useful for protein pairs with few or no similar structures already known. My project focused on this.
Day to day that meant working through ‘tickets’ – tasks the tech team set ourselves each week. Some days that’s writing code, other days it’s reading, making design decisions, running experiments and much more! Every piece of new code that I write gets reviewed by more senior members of the team. That’s been an invaluable opportunity to refine and improve my coding skills and general coding practice.
The biggest learning curve has been learning to estimate how long a task will take me to complete – this is particularly important for tasks that involve skills which are completely new to me. Learning how to break a task down into smaller ones to ensure I’m making progress and still heading in the right direction has been a real game changer to me. One thing that’s helped: TernaryTx doesn’t just score tasks on difficulty, but on uncertainty too. A technically harder task where I already have everything I might need may get less points than a simpler one, where I’m not sure what hiccups might come up along the way.
That approach works because the team makes space for it. Formal meetings are kept short and infrequent, which means people have so much more time and willingness for quick, informal chats whenever something comes up. Andrew, our CTO, was a great help in showing me how to set things up with Amazon Web Services – I had no experience in Cloud Compute before starting at TernaryTx and the documentation for getting started that both he and Naail, Principal ML Research Scientist, had written was exceptionally useful too.
Naail in particular, has been incredibly supportive at each stage of my project. Whether that’s reviewing my ideas for where to take the project or leaving genuinely helpful feedback on my code that’s made me a better programmer. The whole team has been incredibly welcoming throughout, and I’ve thoroughly enjoyed each day.
What’s surprised me most, is how a team of this scale can achieve just by talking to each other! Ternary complexes bring a whole extra set of layers to the design and experimental process compared to a more typical binary interaction. What makes the difference is cross-team collaboration: frequent conversations between teams to discuss progress, challenges, and ideas. Although nobody knows each and every detail of what the other teams are doing, everyone has a clear sense of how their work fits into the bigger picture.
If you’re a student or intern considering this role, apply! This is by far the best experience I could have had of what industry-style research actually looks like – inside a start-up, with a super supportive team and a whole host of interesting research questions and projects. I was given a choice of projects to start with, so there was real room to shape the work myself.
Looking back, my biggest takeaway is simple: good teamwork, communication, and planning can get an enormous amount done in a short space of time. I’m really looking forward to seeing how the TernaryTx team progresses in the next few months.