Thread by @ewanbirney

A personal view point on the #AlphaFold announcement today from the @DeepMind and @emblebi team, part of @embl. TL;DR - I am *still* pinching myself about this.

When @demishassabis and the AlphaFold team first presented the results from CASP to me last November I genuinely almost fell off my chair. I think I swore quite a bit (in a British way) in amazement.

One of the reasons was I knew how rigorous CASP was - 20 years ago people published all sorts of "solving the folding problem" which then... didn't work beyond the training set. CASP cleverly used the fact that there are genuinely unknown structures each year solved by experiment

In CASP you have to put predictions into the competition of targets which are thought to be solved in the year; they cannot be in your training set. The presence of this well run, rigourous competition I think was one reason @DeepMind selected this problem

The other of course is that there is lots of complex data and knowledge - in evolution of protein sequences, in the fundamental chemistry of amino acids, in physical constraints to bring together.

Nevertheless this is clearly not "bring out your standard Inception Neural Network and define your output function" - it is complex to know what to represent to what type of deep learning scheme and how to integrate it. Respect to John Jumper and the AlphaFold team

That was amazing enough but then @demishassabis and the @DeepMind team had to work out the best way to get impact from this - at all sorts of levels - and wisely I think they realised it was best done openly, and if open, go "all in".

Huge *huge* credit here.

So - publish the method; open source the code; run the program on public data, and here it was great that the @DeepMind team approached us @emblebi to partner - in particular Sameer Velankar from @PDBeurope

As @demishassabis has said this is a sort of amplification cycle of the open (experimental) data present in @PDBeurope and @uniprot (the latter based on @ENASequence, part of open DNA) and to open up the results is great and have that part of @emblebi's service resources.

I have just seen the joy on the faces of scientists browsing the data in test mode (I am thinking about @Alexbateman1's face here in particular) and knowing how much information this will provide to scientists everywhere who are interested in structural biology

And the resource being open means that people can download and integrate into their own resources, from the well studied human proteome to, in the future, obscure bacteria and weird protists or viruses... which we might just need to known about in the future!

@emblebi we're ... serious about open data being available for all (globally; academic and commercial) - we allow for time-based embargos for publication, and we work with Data Access committees for honour the consents of human patient/individual data, but no other restrictions.

I know the release of alphafold.ebi.ac.uk will change how protein structure is brought in across science. It is a new tool for structural biologists, alongside their experiments; it is a resource for biologists to dip into structures.

Many many thanks to @demishassabis, John Jumper and the AlphaFold team @DeepMind - for going after a hard problem with no fear; for holding yourself to a high standard of science; for making the results open. You rock.