Thoughts on the #AlphaFold #Deepmind 'we done the proteome' news:
(1) This is great.
(2) This was always going to happen. I'm surprised they did it this fast, but cool.
(3) The dataset will be invaluable for hypothesis generation.
(4) Hypotheses will still need to be proven at the lab bench.
(5) This will advance structural biology, allowing phasing of MX datasets and tracing of domains/proteins into EM maps.
(6) I'm interested to see how their models perform with drug & biologic design.
#AlphaFold
(7) I'm interested to see how their model aid in construct design. Knowing disordered/unstructured regions is super important for designing well behaved, soluble recombinant proteins for structural study. #Alphafold
(8) I'm also interested to see how their models perform in multidomain and multisubunit complexes. These are often the engines of biochemistry (eg ribosome) and understanding the composition and dynamism of these complexes is key to understanding molecular biology #AlphaFold
(9) Given the press release driven nature of #Alphafold #Deepmind's work to date, I'm sure we will see a few breathless articles proclaiming that structural biology is done. Such takes are rooted in 1990s-era structural biology where the structure was the story.
It's rare to publish a structure and nothing else these days. Now, the structure informs the biology, and it is the biology that is the story. That biology still needs to be tested, in cells or test tubes, biochemically or biophysically.
But anyway this is really cool. I look forwards to playing with the models of my favourite proteins, and maybe even phasing a few old datasets I have lying around.
(this still makes me twitchy.)
(From nature.com/articles/s4158… )
Now I think that they're guilty of a bit of overreach and a bit of over-enthusiasm.

The method doesn't work for RTKs that I've seen so far. Uniprot has annotations about what's intracellular, what's extracellular, what is a signal sequence, etc
Why not use that metadata to trim your models accordingly? Do separate models for intra- and extra- cellular domains? Not difficult.
Another one. I did the structure of the blue bit in 2016. The red stuff is unstructured. It just loops and whirls around, and there's a TM helix.
LAMB2 doesn't look like this. It's clear #Alphafold deals very well with individual folding units, but not so well with long-range, inter-domain stuff. I won't be hanging up my pipettes just yet.
Anyway, here's the paper
nature.com/articles/s4158…
Early #Alphafold feedback from my contacts:

As well as a few models that are biologically implausible, a couple of cases where Alphafold models have suggested why certain constructs don't fold/express and a case where an inter-domain contact might give functional clues.
See - we're generating hypotheses with the #Alphafold dataset already!
Sorry I've been quiet about this, this evening. Thing is I'm busy collecting crystallographic data @Diamond_MX

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Dave Briggs

Dave Briggs Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @xtaldave

22 Jul
All predictions are high accuracy, some predictions may be more high accuracy than others. 😬
He's an example of our favourite RTK, RET.
The signal sequence is still attached (red top left), and the Kinase domain (inside the cell) nestles up inside the curve of the CLD domains (outside the cell). The TM helix is at the bottom.
Read 5 tweets
22 Jul
Someone should tell the gubmint that relaxing restrictions is irrelevant with over a million active cases* of Covid, and a further proportion of the population self isolating - the economy ain't going to spring back into action any time soon.
Half my trains are cancelled due to lack of available train crew. Folks can't get to work, or they have to crowd on the trains that are running, further increasing the chance of exposure. Just as mask rules are relaxed.
*
worldometers.info/coronavirus/#c…

UK active cases count accurate when posted.
Read 5 tweets
17 Jan
"I know Kung-Fu"
"Remember the tooth"
"Make it so"
Read 5 tweets
15 Jan
Woah - in the Crick testing pipeline, 10% of positive cases were the B.1.1.7 variant in the first week of December.

In the first week of January, it was 90%.

That seems crazy fast.
For context, this was relayed to staff by management in a "town hall" today.

Whilst some samples will have been sequenced, these values were determined by the absence of the Spike signal in the Thermo 3-gene PCR test.
Certainly, this data suggests that B.1.1.7 has a significant transmission advantage.
Whether that's down to enhanced ACE2 binding, or bigger viral loads, or some combination of the two, remains to be proven, I believe.
Read 5 tweets
1 Dec 20
Right, I know I keep banging on about Deepmind & Alphafold, but something I (a) do for a living and (b) am knowledgable about, is finally in the news, so suck it up.

*deep breath*
It's not available to all - the toys/source are (to the best of my knowledge) not available to other researchers.

So it isn't actually advancing science (yet). To do that it needs to be open.
And it is built on open depositories of protein structures (PDB) and sequences (Uniprot or equivalent), and on academic pieces of software (HHblits is in the mix, for example, I believe)
Read 7 tweets
30 Nov 20
Not to undermine this, and the work they are doing is excellent,l & important, but they did claim this at the last CASP.

Caveats then were that it only works on small proteins/domains in isolation, which is still super useful, but not what a lot of structural biologists do.
I await the publication with interest.
An RMSD of 1.6Å is damn good homology model.
deepmind.com/blog/article/a…
Read 6 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Follow Us on Twitter!

:(