DeepMind AlphaFold

DeepMind, an AI lab & complete outsider to the field of molecular biology, beat top pharmaceutical companies with 100K+ employees like Pfizer, Novartis, etc. at predicting protein structures. This is huge! DeepMind didn't yet release the paper or code, so I browsed the CASP reports and different related papers to reverse engineer the architecture as best as I could. What I ended up doing is forking a related research papers code and repurposing it, since its very similar (2 residual networks were used). In this episode, i explain the different components of its architecture, why the protein folding problem is so important, and give some programmatic examples. Enjoy!

Code for this video (with coding challenge):

Please Subscribe! And Like. And comment. Thats what keeps me going.

Want more education? Connect with me here:

More learning resources:

Join us at the School of AI:

Join us in the Wizards Slack channel:

Please support me on Patreon:

Signup for my newsletter for exciting updates in the field of AI:

[mizuhonova]

How to get a paper published in 2019.
1. Find a problem
2. Throw AI at it

[motog]

Attention Biology students !!!
Now you guys don't have any excuse to escape the AI route 😀
Thanks again Siraj 👍❤️

[letsoverengineer]

the quality of your content increases video by video

[mkmabdelkawy]

I was just searching about AI/ML protein folding and found Alphafold in an article for most recent progress in that topic, then googled it and found you just making a video about it 18 hours ago. You are rocking on every topic Siraj, keep it up.

[raunakdoesdev]

"I went ahead and recreated it programmatically"... only changes the README on his fork.

[cu]

I'm still trying out Ordinary Differential neural network and he released another video

[einemailadressenbesitzerei]

This is really a great Video on a High Level scale. That was high quality and so inspiring. I wish there were more of them again.

[qncsc]

_" ... to your moneymaker."_
LOL, i lost my focus at that point
always the hidden "sirajism"

[Soulthym]

It's been a few years that we talk about using deep learning for protein folding, and it's finally here! I can't wait to see the next version of it, it reminds me of how deep learning took over image classification.

[tylerk]

This is one of those rare AI results where the response isn't just cutting edge AI, it's also something that is directly and immediately applicable to a critical real world problem. I look forward to your deep dive when the code is released.

[soumyakundu]

Great Video! Really simplified the blog and the paper! Just wanted to mention one point, the team consisted of members with the knowledge of structural biology which is a subdivision of molecular biology and biophysics. So technically they weren't noobs at all but experts in it! Matter of fact, to quote "Structural biology is a branch of molecular biology, biochemistry, and biophysics concerned with the molecular structure of biological macromolecules (especially proteins, made up of amino acids, RNA or DNA, made up of nucleotides, membranes, made up of lipids) how they acquire the structures they have, and how alterations in their structures affect their function."

[jessedampare]

9:02 got me deeeaad🤣🤣😂 shoutout to Microsoft Lmaooo

[TheNeilDarby]

Hopefully in the future we can use a complete understanding of protein folding to re-engineer photosynthesis so that we can finally have a scalable negative emissions technology.
Just might save the world.

[meisterhubert]

Hey Siraj,
I studied physics but you motivated me to look deeper into ML. Now I start my PhD to work on acoustics with ML wizardry :)

[sprocketus]

That's remarkable results, thanks for video.
P.S. I'm very curious about applying AI to code generating, can you please highlight these theme)

[Chesstiger]

Both the content and the presentation is on point!

[AbhishekKumar-yvih]

Waiting for the next generation of AlphaFold !!

[Islam_first]

Siraj!
I am almost done with ML course by Prof Andrew Ng!! What's next?
I know you have Deep Learning in 6 weeks, Data Science in 3 Months, ML in 3 months, and I am also very interested in Deep Learning on Coursera by same Professor.
Just trying to understand what I should tackle next!

I wish you good health and great success in AI and beyond!

[merlinthelemurian]

bro wow, it's been like 8 months since i've watched your videos, the level of improvement is insane! i am extremely impressed.

[blakeallenvideo]

Love the video! One thing I wanted to make note of @siraj, Rosetta is actually a suite of protein folding tools and structure predictions (not a visualization tool like pymol for example). So DeepMind actually ran their output through a series of additional statistical systems (hidden markov models) that further predicted the ideal 3d conformations. Perhaps a missing step in this process!