the more I read about the #alphafold2 details, the cooler it gets. This is not your basic machine learning, but incorporates a ton of domain expertise. Seeing how far deep learning is, I realize it embodies what I had in mind during my PhD with molecular chemometrics :)
the first generation DL applications where just throwing a lot of data at it, but #alphafold2 actually shows it has no problem with embedding prior knowledge... the attention/transformer aspect (think of it like variable selection) is pretty awesome.
besides intelligent variable selection, you will also find traces of idea like kernels (like those in SVMs) and domain-driven measures of similarity.
After my PhD, the biggest bottleneck, in chemistry, was basically the amount of data. Or lack of that. #alphafold2 has the advantage of decade old #openscience practices, and a flood of mmCIF files. Basically: there is enough data to train many parameters.
and that gives you a playing field to explore how to put the domain knowledge into the model, and particularly if at all. I cannot stress enough how awesome the #alphafold2 story is. Congrats to all involved!

• • •

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

Keep Current with Egon Willighⓐgen

Egon Willighⓐgen 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!


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 @egonwillighagen

6 Aug
"key performance indicators"

when not properly defining what "performance" is, they will become as useless for research as journal impact factors.

So, probably worth reading what they came up with as KPIs.
so, what @eoscassociation KPIs can we expect? the two cited documents report the following KPIs (and I'll comment) 1/
@eoscassociation KPI #1: Percentage of publications [..] that become immediate
open access: 70%

-> not very ambitious, but better than nothing. All colors allowed, it seems. 2/
Read 17 tweets
24 Jun
the Eclipse-based #bioclipse GUI has not seen releases after the 2.6.2 version, and like Eclipse software, building it has been difficult. But I was still using Bioclipse a lot and updated to libraries became doable on my laptop, but ... 1/
for a while I supported people in my group to set up a post-2.6.2 Bioclipse from the source code repository. That allowed me to write new plugins, etc. I even had figured out how to set up 2.6.2 update sites for new functionality 2/
I realized I needed something else, something less time consuming. Oh, and I also wanted to be able to run Bioclipse scripts on the command line, without a GUI. So, I had this idea: Maven, command line, Sonatype, Groovy Grab for easy install. 3/
Read 17 tweets
16 Apr
updating slides for my contribution to the @lipidmaps Training School...

Wikidata currently has 18201 LIPID MAP identifiers, 3434 are associated with CAS RNs confirmed by @CASChemistry Common Chemistry
@lipidmaps @CASChemistry a bit more about this... attached is the slide I presented at the LIPID MAPS and MAP/UK Training School last week 18201 lipids: 3...
@lipidmaps @CASChemistry but this was still mostly based on earlier work, and overnight another @wikidata job ran to add ~3400 @lipidmaps identifiers, based on existing chem structures and InChIKey match, so let me post the updated numbers:
Read 11 tweets
14 Apr
I am editor-in-chief of @jcheminf, a journal where we push for more #openscience in cheminformatics (and via this, in chemistry). All articles are CC-BY. This is what I see when I visit the website: Nothing, except a page covering "Your privacy" pop
@jcheminf I do not seem able to decline cookies. Isn't that a violation of EU law?
@jcheminf It even hides what the "stricly necessary cookies" are. Apperntly, this includes cookies from (which sounds like a Sesamestreet domain, but proably isn't)
Read 8 tweets
12 Apr
2nd speaker in this #ACSSpring2021 session is @CameronNeylon, going to put the ACS and RSC in the ring to face each other
@CameronNeylon CN brings up that the two societies are major publishers but also have their own national policies #ACSSpring2021
Read 4 tweets
12 Apr
#ACSSpring2021 session on "innovations in open data exchange and information dissemination has started". First speaker is Allison Leung on @ChemRxiv
@ChemRxiv Leung: "why use preprint? speed, feedback, credit, visibility" #ACSSpring2021
@ChemRxiv Leung: "13 thousand submissions from 90+ countries; 8,200 preprints posted; 25 M views and downloads" #ACSSpring2021
Read 7 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!