Disappointed that #AlphaFold didn’t get references quite right- first washttps://journals.plos.org/plosone/article?id=10.1371/journal.pone.0028766 …
-
Prikaži ovu nit
-
Odgovor korisniku/ci @deboramarks
Is key ingredient in
#alphafold evolutionary couplings ? As in first successful 3D folding http://bit.ly/EVfold1 reply 3 proslijeđena tweeta 4 korisnika označavaju da im se sviđa -
Odgovor korisnicima @sandercbio @deboramarks
Coevolution information sure is important, but for many CASP13 hard targets, I guess direct information is not much better than mutual information when deep ResNet is used.
1 reply 0 proslijeđenih tweetova 4 korisnika označavaju da im se sviđa -
Odgovor korisnicima @jinboxu_chicago @sandercbio
Testable?
#alphafold relies dependencies across evolution - simple. Only other way is real physics.0 proslijeđenih tweetova 1 korisnik označava da mu se sviđa -
Odgovor korisnicima @deboramarks @sandercbio
A while ago, I did some test on the CASP12 hard targets. Please check out some results in the 2nd paragraph of subsection "Dependency on Multiple Sequence Alignment and Direct Coupling Score" in https://www.pnas.org/content/116/34/16856#F2 … . The results were based upon a deep ResNet of 60 conv layers
1 reply 0 proslijeđenih tweetova 6 korisnika označava da im se sviđa -
Thanks. I'm surprised that there is no analysis of the effect of the depth of the alignment and generalisability to proteins with fewer homologues. How much does the method learn something general and how much is based on a more robust extraction of dist distributions from MSAs?
1 reply 2 proslijeđena tweeta 2 korisnika označavaju da im se sviđa -
Odgovor korisnicima @LindorffLarsen @jinboxu_chicago i sljedećem broju korisnika:
First part of your question is somewhat addressed in panel A of our paperpic.twitter.com/nw6GafIOSZ
1 reply 0 proslijeđenih tweetova 3 korisnika označavaju da im se sviđa -
Odgovor korisnicima @labriataphd @jinboxu_chicago i sljedećem broju korisnika:
Nice, thanks. Though it's still unclear how (statically) significant the improvement at low Neff is btw 12 and 13. I'd think it would be much cleaner if single methods and targets were evaluated at different (artificial) depths. I'm sure this analysis has been done.
0 proslijeđenih tweetova 2 korisnika označavaju da im se sviđa -
Odgovor korisnicima @LindorffLarsen @labriataphd i sljedećem broju korisnika:
Yep! For each protein family, as you increase number of sequences so does the accuracy. But it is definitely "relative". For some protein families, you can get away with a single sequence, others you need thousands for same accuracy.
0 proslijeđenih tweetova 4 korisnika označavaju da im se sviđa -
Odgovor korisnicima @sokrypton @LindorffLarsen i sljedećem broju korisnika:
The challenge in CASP is that many hard targets have very few sequence homologs.
1 reply 0 proslijeđenih tweetova 3 korisnika označavaju da im se sviđa
This is also true for a lot of proteins outside CASP. In fact for many of the interesting proteins we work on. One may be fooled by the fact that the PDB is biased towards proteins with more homologues than a typical protein as @wim_vranken showedhttps://www.nature.com/articles/srep36679 …
-
-
Odgovor korisnicima @LindorffLarsen @jinboxu_chicago i sljedećem broju korisnika:
…and that are relatively easy to crystallize, and that someone has been able to convince a finder are important for one reason or another…
0 replies 0 proslijeđenih tweetova 0 korisnika označava da im se sviđaHvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.
