Re: DeepMind’s AI Advanced Protein Structure Prediction tool Open Sourced
Hi Steffen,
On 2021-07-18 21:47, Steffen Möller wrote:
> Following the references in
>
> https://www.nature.com/articles/d41586-021-01968-y
>
> I found this reference
>
> https://github.com/RosettaCommons/RoseTTAFold/tags
>
> but I admittedly cannot tell that I'd have fully grasped who is doing
> what and publishes what software where, yet.
This would definitely be great to have in Debian. I expect this protein
structure predictor heavily depends on pre-trained neural networks.
Making them DFSG-compliant might be very difficult (see Unofficial
Policy for Debian & Machine Learning [1]).
> The context:
>
> The biochemistry happens in 3D as proteins (and some functional RNA),
> while all we easily get are 1D DNA sequence data from which one can
> predict the sequences of amino acids that form the protein. To get from
> that polymer of amino acids to the 3D structure one typically looks for
> patterns of sequences that have been observed in structures that have
> already been determined. Or one try computing that structure de novo,
> i.e. without a template, and ... wait.
>
> Once we have the protein structures one can better understand the effect
> of mutations, and start other sorts of simulations, like disturbing
> protein interactions with compounds, i.e. find new drugs.
I am particularly interested in this. However, this field is quite
unrepresented in Debian, AFAIK. Some time ago I packaged
MacroMoleculeBuilder [2], which does homology modeling. All I have tried
is a couple of examples, but it is definitely worth giving a look.
Currently I am looking into ProMod3 [3], which seems to be the engine
behind the great SWISS-MODEL service [4]. I seem to have figured out the
dependencies, will go on to packaging next.
[1] https://salsa.debian.org/deeplearning-team/ml-policy
[2] https://simtk.org/projects/rnatoolbox
[3] https://www.openstructure.org/promod3
[4] https://swissmodel.expasy.org/interactive
Best wishes,
Andrius
Reply to: