Tristan Free

speaker

289 appearances 6 recordings 1 series first heard Jun 2026 last heard 23 Jul

Tristan Free’s voice in public audio — every appearance, attributed to the second.

Trend

recordings per month · last 12 months

Recordings per month over the last 12 months — 6 in all, peaking in Jun 2026 with 4.

Appearances

newest first · ▶ plays the moment

Talking Techniques · AI & Antibodies miniseries | Nanobody thermostability prediction and the great data challenge · 6 Jul 2026 podcast

Um and so in your paper you addressed the limitation of AI that reaches far beyond the life sciences and and antibodies, um, which is the quality and size of the data it's trained on.

So as those listening to this podcast likely work in highly specialized fields, this is an even more pronounced problem, right?

You have less data availability than you would for sort of more general applications.

Can you uh can you tell us how you set out to develop a useful model with a relatively small data set?

So how did you kind of try approaching this situation?

Fantastic.

And and what kind of accuracy did you manage to achieve with that model?

No, fantastic.

And and so you you mentioned it had a a kind of error margin of about four or five degrees.

Um is that, would you say at this point narrow enough to be sort of truly useful in that early nanobody design phase?

Or is it kind of that's a first stepping stone on to getting towards something that is?

Fantastic.

Um and then so one of the things you found during the the sort of course of that paper as well is that you had a a kind of general protein model that we you were using to assess this, and then you also had the specialized antibody models that you were kind of comparing against.

And that general protein model outperformed the specialized antibody ones.

W why do you think that was?

Uh that kind of leads me on to my next question, which is i i how much do you think the model that you did develop could be improved with more data and how much more would you need to see a a a serious step change in the um the ability of it?

So it it sounds like it's more about um so whilst there is, you know, that that sort of it hits that plateau of improvement around one thousand sequences and then after that point what's really going to be valuable is is characterizing that data and and and also um

manipulating your model or sort of training telling your model what it is that it needs to be focusing on, what the core things are, so that it's not just, as you mentioned earlier, looking at binding properties and sort of key things that's classic to to antibodies, but it's really looking at all these different relationships and getting a better sense there.

Yeah, exactly.

Showing 41–60 of 289 · page 3 of 15