Aubin Ramon

speaker

346 appearances 1 recordings 1 series first heard Jul 2026 last heard 6 Jul

Aubin Ramon’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 — 1 in all, peaking in Jul 2026 with 1.

Appearances

newest first · ▶ plays the moment

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

So maybe it's just that it's just because the antibody specific one are small models.

Um, not has hasn't seen that many sequences.

Um and you just see maybe like a big model.

Yeah, it's the eternal question of uh do we need more m more data and more and more data.

This is what we hear in all the conferences.

Um so actually we did a preliminary study in an OML paper where we generated this data set of um artificially generated biophysical properties, which is in that case uh solubility, but we computed with CAMSO, a tool that my PI developed during its PhD.

So CAMSO is a simple linear combination of features from a given sequence, so nothing um mash learning related.

So we have then this unlimited amount of data.

Artificial data, thanks to that.

And I train the same architecture nanometer to predict this artificial stability by given data set sizes.

So I went from ten thousand sequences down to ten, I think.

And what we saw is that

as expected, the less data you have at some point is gonna drop.

So the training performance is always gonna be good, but then the test performance is gonna be extremely bad.

This is what we call the fitting.

And we have this at around

One thousand.

So it's not

So the curves you can see them in supplementary.

Showing 161–180 of 346 · page 9 of 18