AlphaFold and the Grand Challenge to solve protein folding

If you want to support this channel, here is my patreon link:
/ arxivinsights --- You are amazing!! ;)
If you have questions you would like to discuss with me personally, you can book a 1-on-1 video call through Pensight: pensight.com/x/xander-steenbr...
--------------------------------
AlphaFold is DeepMinds latest breakthrough addressing the protein folding problem. Using an advanced Deep Learning architecture that achieves end-to-end learning of protein structures, this work is arguably one of the most influential papers of this decade and is likely to spark enormous advanced in computational biology and protein design. This video covers the entire architecture of the model as well as training principles that led to the incredible results of AlphaFold2!
AlphaFold Nature paper: www.nature.com/articles/s4158...
AlphaFold Codebase: github.com/deepmind/alphafold
Work from the Baker lab: www.bakerlab.org/
Fabian Fuchs' amazing blog on equivariance: fabianfuchsml.github.io/alpha...
Ongoing Open Source effort to reproduce AlphaFold: github.com/lucidrains/alphafold2
::Chapters::
00:00 Intro
02:28 The Protein Folding Problem
05:29 AlphaFold1 revisited
06:10 Multiple Sequence Alignments (MSA)
08:10 Distograms
12:29 AlphaFold2
14:52 The Evoformer
19:07 The Structure Module
28:13 Zooming out: looking at the future

[@FarrukhTickets]

Where do you go man!!! We need you to wrap our heads around Chat GPT 3,45,6,7,8,9,10... Not to mention, Bard, LamBda, LLama and others. Come back to your channel.

[@julian78W]

The sheer effort you put in your videos is mind blowing.

[@memebandit]

I can stop it do it

[@InturnetHaetMachine]

I just love that you actually go into the inner workings and important details of these papers in a coherent manner, and not just reading paper line by line and just repeating the information. Thanks for starting to make these videos again.

[@joankim123]

I love how this explains the reasoning behind all the steps. Most content just explains what and how, not why

[@mohamedayoob4699]

A simple man here. Xander posts a video and I click it.

[@navneetbung9609]

Thank you for explaining a complex architecture in such simple terms. Great work!

[@yarrowflower]

Wow, you're uploading again! Great to see you!

[@linzhang4154]

Probably the best AlphaFold2 explained!

[@johnf9507]

Gives a great overview before attempting to read the paper. Saved me a lot of work. Many thanks.

[@vadimavkhimenia5806]

Please continue making videos. Your channel is great.

[@UnpluggedPerformance]

Hey Xander, Toffe video man!! Leuk om iemand vanuit Belgie tegen te komen in de online data science en machine learning world! Ga zo door en super bedankt voor je inzet voor deze mooie video!!

[@mahzabeenkamrulsuha4124]

Tears coming to my eyes! thank you so much. have to present the paper soon and got this... soooo many new terms for me as I have no idea about AI, ML, advanced computer language knowledge and this is helping so much I can't say. Haven't finished watching yet but going well. New subscriber. 😊

[@anishahandique4815]

After going through most of the KZbin videos on this topic. This one was one of the best out of all. Very clear and crisp explanation. Thank you ❤

[@jucls2002]

I love it :) So happy that channel exists

[@erv993]

Great video! Thanks for breaking down to the very details, it's indeed mind-blowing

[@samuelelwell7575]

Superb video, best one on AlphaFold 2 I have encountered by far!

[@victormontalblancafort8473]

I just really enjoyed the video! Both the video perse, and the conntent was wonderful! Keep doing them :)

[@alexis3620]

Great Work ! Your video is a real effort to give accessible the meaning of this paper !

[@yaiirable]

Thanks for all that detail! It would have been interesting to hear a bit more about the potential applications of this technology at the start of the video

[@ajaypeddada2981]

glad to see you back :D

[@CAME18471]

One of the best a science communication KZbin channels ever!

[@sergiugopleac]

Great video! Congratulations 👏

[@peterzhung3153]

Thank you for making this amazing video!!

[@finnweikert3430]

thank you sir! appreciate the effort that went into this video

[@DjSapsan]

What a time to be alive!

[@Bjarkediedrage]

2.0

[@alessandraneis6692]

best explanation of alphafold!!! congrats and thank you!!!

[@hongaivu1700]

The most educational informative video that I've ever watched on KZbin, bravo! I wonder what is your background?

[@AkashMishra23]

My dude is back, Amazing and well made video, Absolutely love watching your content

[@robertc6343]

I’m saying the same thing! Nice to see him back.

[@antoniojosepretomartinsgom9503]

Very cool explanation with respect towards the scientific community's efforts as a whole

[@sandrallancherosg]

The video I dreamed about!! Thanks!!

[@infoman6500]

Very interesting. It looks like Nature is alive -very much alive.

[@salmagamal5676]

Incredible work!

[@aszyon5962]

Still the best on AlphaFold I’ve listened to

[@mohamed_bouallegue]

Excellent videos as always ♥

[@arjangvt]

Excellent video!

[@davidliu5075]

Very informative! Thank you so much!

[@kennethverstraete619]

Great presentation, thanks!

[@yusun5722]

Excellent explanation.

[@danalin8574]

This is so good! Thank you!

[@epadam]

Best KZbin Channel ever.

[@patriciobarletta8549]

Quick correction: AF2 deals with the protein structure prediction problem, not the protein folding problem.

[@ArxivInsights]

Aah yes you're correct, my bad! Too bad KZbin won't let you edit an existing video...

[@mipsee5967]

What is the difference?

[@antiregime88]

@@mipsee5967 essentially "protein folding" means the whole process from a linear aminoacidic sequence to the final 3D structure. This involves a lot of intermediate states that cannot be predicted by AF2. Instead, "Structure prediction problem" is the challenge to predict only the final and functional 3D structure knowing the aminoacidic sequence, that's what AF2 does.

[@mipsee5967]

@@antiregime88 Oh I see, thanks a lot man

[@SC-ib4oo]

Excellent video! Thank you very much

[@learnfromIITguy]

Amazing presentation

[@aportorak8575]

Just found your channel. Amazing stuff

[@urfinjus378]

Thanks for the video!

[@park8919]

deeply appreciate your work from japan

[@slee9254]

Amazing video. Thank you

[@iot3136]

Amazing. Thank you 👏♥️

[@kaiz6997]

extremely amazing, thanks for creating this incredible vedio

[@oscarrojasl]

Great content! Thank you

[@hom01]

great video ❤️

[@iamr0b0tx]

where have you been? It is nice to see you again

[@FernandoOliveira-kt3eh]

Nice! Vid!! It would be perfect with some form of "concept indexing".

[@rajanmoodley4822]

This is such an informative video. Thanks Xander. When are you doing a video on MuZero and AlphaTensor

[@KP-ed4wm]

This is so useful, thank you! A little confused about what a pair representation actually is though, still. Is each i,j value the average distance between any residues i and j found within any of the known sequence-structure proteins from PDB? Or is it something else?

[@shrutigupta8166]

I like your videos, please make a video on meta learning:social cognition and consciousness in brain and machines

[@lolerie]

You are right about Leela Chess, it did destroy Stockfish but then stockfish became a NN too and nore or less destroyed Leela! And yes with Go, NN is the best.

[@fraune]

The original machine had a base plate of prefabulated amulite, surmounted by a malleable logarithmic casing in such a way that the two spurving bearings were in a direct line with the panametric fan.

[@Diagnoc]

Fascinating stuff, although you got me lost after ten minutes or so (I don’t have a clue about programming).

[@ashtonreason3444]

Layman trying to figure this out. Im lost but goddam this is amazing

[@Rafagenetico]

Great explanation. Is Alphafold2 recommended to modeling small peptides (5-40 aa)?

[@C01A60]

Why has this amazing channel stopped making new videos?

[@JorgeRamos-xw6dy]

I admire deep mind but hope that Google recognized and compensated all parties involved.

[@kaaajeee]

6:40 - 7:20 where is that amination from, pretty cool. would like to see the full movie.

[@courtneyshearer3796]

Hi Xander, what software are you using to look at MSA?

[@davefar2964]

Thanks a lot for this very helpful video. I did not know that they used quaternions. The backbone frame input of the structure module has the shape (r, 3x3) and (r, 3), which I interpreted as the rotations and translations of each frame. With quaternions being used, what does the backbone frame input of shape (r, 3x3) and (r, 3) contain?

[@Anujkumar-my1wi]

can you explain the concept of 'state' used in reinforcement learning as there are lot of misunderstanding regarding its defintion

[@luxliquidlumenvideoproduct5425]

One must stress what you say at the end of the video at 28:20, that although AlohaFold 2.0 can predict native confirmation of an amino acid sequence, there are other contributing factors, and the algorithm isn’t able to answer the why, nor how proteins find their native state out of the vast combinatorial complexity of native confrontation structures. Levinthal’s Paradox.

[@p0tmuffin69]

So smart

[@CaribouDataScience]

What kind of hardware is alpha2 running on?

[@DanKolis]

What about noticing John Moult has spend his ENTIRE LIFE making Casp-XX exist ? None of this figuring out would have happened without the insight and attention to detail, year after year to make a system for thinking about it all exist !

[@bharasiva96]

Man lucidrains is the G.

[@ovechristophermattmann880]

Great stuff! But maybe place your teleprompter somewhere behind the camera 😉

[@AChadi-ug9pg]

❤️

[@W0954]

I don't understand the not biology part, what should I study to be able to understand it ? Computer science ? Statistics ? Mathematics ? I have a biotechnology degree and I don't understand how AI works.

[@bighands69]

Protein folding has not been solved but they now created the pick axe and hammer to dig the mine.

[@winll5200]

now i really think my IQ had leveled up

[@sofeaarina8506]

what is pair representation?

[@Spiritusp]

I keep hearing "EvilFormer"...

[@aszyon5962]

😀😀 me too

[@renegroulx7029]

Im not really getting it.

[@dororo2597]

"Considered solved ?" Isn't it still didn't solved yet

[@arielsioz5036]

TL;DR ACT WHAT YOU PREACH, Xander never even open sourced his "fun side-project", Synesthesia, and is now calling out Deepmind for being late into open sourcing their phenomenal research. Most entertaining thing is that Xander ( the dude in the video ) couldn't even help himself from open sourcing one of his "fun side-project" ( as he called it ), Neural Synesthesia, but chose to turn it into something profitable. Nothing wrong with this, it only gets me when he *criticizes* the actual brightest people out there who came up with AlphaFold and potentially wanted to monetize it.

[@ChazyK]

I didnt know Justin Bieber understands machine learning so well.

[@TYFILMPRODUCTIONS]

You mean James Franco?

[@user-or7ji5hv8y]

Are most cryptography in cryptocurrency based on the assumption that it would also take the age of the universe to break it?

[@DavidHaselbach]

We don’t quite consider protein folding to be solved though

[@lolerie]

It is now solved. Just look into more recent twitter posts. It IS solved.

[@jameshammond2846]

@@lolerie it clearly works well with extant protein structures, but has anyone yet expressed completely artificial proteins and compared the ezperimentally determined structures of these to the alphafold predictions?

[@lolerie]

@@jameshammond2846 yes. Proteins from Foldit. Also someone just checked very new PDB structure, it works too.

[@jameshammond2846]

@@lolerie how do you mean, proteins from foldit? Are the proteins on foldit from trully random sequences?

[@jameshammond2846]

@@lolerie and working on new PDB structures is impressive, but did these structures have high homology with other structures in the PDB?

[@kalipogmail8217]

if this is early AI we are F**d

[@yakisalosa1080]

Those moustaches man, can't focus on what you're saying.

[@lolerie]

"Proteins are everwhere" No, proteins are constructed by ribosomes, which have RNA as active center, i.e. proteins are being contructed by ribozymes.

[@ArxivInsights]

That's exactly what my previous video was about!

[@adelarscheidt]

Overcomplicated explanation

[@Sodomantis]

Oh fold. I thought it said Alpha Food.

[@antoniogarest7516]

Great video!

[@jijie133]

Great video!