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

There are interactions between any sh within any chains of the proteins, PDB, the sequences, the um large language model have have seen.

So maybe by having um seen more examples of this hydrophobic interaction, hydrophobic interactions, um you might give features to the our nanomel model that are a bit more meaningful.

And then down the line.

um be um more accurate.

So this is what we thought.

And maybe because

If you have a antibody specific language model, it might have focuses learning on antibody only features.

I don't know, such as like binding, for instance, or like binding pattern in the sequences that you don't really see in proteins.

Um and then he he didn't really like spend much time into learning the features important for stability.

Um so that's the first intuition.

But then we can I can place my own Devos advocate and say, um yeah, but um when I use a protein general thermostability predictor like deep stab B, but it's completely lost on antibodies.

So he

seen much more sequences by antibodies.

So you might have learned also like all these good interactions and uh correlation but you need for predicting stability.

But actually on spaces like antibodies it was lost.

So maybe the answer to that question is not that deep.

It may be just because the protein language models I use, like ESM are massive, they have a lot of parameters and been trained extensively.

the representation they give are bigger than the representation they give, but antibody specific give.

So you are you have more data move mo sorry, more features um to give to your model down the line uh for for uh transfer learning.

Showing 141–160 of 346 · page 8 of 18