Davide Sala

Computational Structural Biologist. Institute for Drug Discovery, Leipzig University. MeilerLab

[Davide Sala @sala_davide]

@AdrianoAguzzi @chribo99 That’s the point, it depends on how large the system you want to predict is. For instance, In cofolding 80gb may not be enough for systems >2k residues. That limit drops with recent models leveraging structural tokens.

[Kevin K. Yang 楊凱筌 @KevinKaichuang]

In 1965, Margaret Dayhoff published the Atlas of Protein Sequence and Structure, which collated the 65 proteins whose amino acid sequences were then known.

Inspired by that Atlas, today we are releasing the Dayhoff Atlas of protein sequence data and protein language models.

[Davide Sala @sala_davide]

@DdelAlamo I liked a farewell to entropy. Linking information theory to stat mech and thermod

[Davide Sala @sala_davide]

@DdelAlamo Just include water in AF3 predictions

[Davide Sala @sala_davide]

@DdelAlamo Yeah, you're probably right. Training set is prob the most relevant bias

[Davide Sala @sala_davide]

@DdelAlamo Populations are affected by external factors. Proteins may be more or less sensitive to external factors. Maybe lrmp is more sensitive than those soluble proteins to ext perturbations. I.e. the dynamics they were checking is encoded in the MSA while for lrmp it doesn't.

[Davide Sala @sala_davide]

@huhlim With a very shallow random MSA I got a lot of broken structures due to missing information in both template and sequences. Keeping only seqs > 50% identity may have helped

[Davide Sala @sala_davide]

Dear all, if you're interested in the topic we wrote a review on modeling different protein conformations with alphafold. @MeilerLab @Mchaourablab @fengel97
https://t.co/OjyahW6T3r

[Davide Sala @sala_davide]

@ferruz_noelia I agree. We should have a European legislation to treat that matter.

[Davide Sala @sala_davide]

@RolandDunbrack They look great!

[Davide Sala @sala_davide]

In this work, we exploited two APIs (GPCRdb and KLIFS) to bias the prediction of desired conformational features of GPCRs and kinases with AF2.
Main results in 3 points:
https://t.co/GMmlIgM1wc

[Davide Sala @sala_davide]

Suggestion: try different MSA depths (8:16, 16:32, 32:64 ...). Evaluated structural and confidence scores to find the perfect balance that does not change the desired structural bias and maximize accuracy.
Code works with ColabFold and can be found at https://t.co/eML0J1pYam

[Davide Sala @sala_davide]

3. On kinases, predicted ensembles were enriched with user-defined structural features. We encoded DFG in/out/all, Ac_h in/out/all, and salt bridge K17-E34 yes/no/all so far. An API with better conf annotations could do better. Through @RolandDunbrack classification for instance

[Davide Sala @sala_davide]

intermediate states mainly responsible for coupling selectively.
https://t.co/yFHrwMn1WR

[Davide Sala @sala_davide]

We've written an Opinion piece on AlphaFold2 conformations for SBDD. GPCRs focused. As pointed out by @BonomiMax a couple of days ago, we are already in the 'protein conformations era'. Are these suitable for SBDD? @MeilerLab and co.
https://t.co/pWMYfbHtuj

[Davide Sala @sala_davide]

@PostDocsForum Never. I tried to rephrase a multi paragraphs text and I had the feeling that was less accurate. But I never asked to recap rather than to rephrase, and it's not the same thing.