Ri on AI: Understanding AlphaFold - with Dame Janet Thornton

We’re exploring how AI is impacting different areas of scientific research, and in this episode we discuss AlphaFold, the machine learning programme from DeepMind that can determine a protein structure from its amino acids alone
Lisa Derry is joined by Dame Janet Thornton, an oracle of bioinformatics and former Director of the European Bioinformatics Institute, to discuss how AlphaFold has revolutionised its field, and what it could do in the future.
This is the full video version of our latest podcast episode - let us know what you think of the new video format!
Production: Lia Hale
Thank you to our Christmas Lectures supporters and our title partner, CGI.
Search the AlphaFold database: alphafold.ebi.ac.uk/
Find out more about the 2023 CHRISTMAS LECTURES: www.rigb.org/christmas-lectur...
The Ri is on Twitter: / ri_science
and Facebook: / royalinstitution
and TikTok: / ri_science
Listen to the Ri podcast: podcasters.spotify.com/pod/sh...
Donate to the RI and help us bring you more lectures: www.rigb.org/support-us/donat...
Our editorial policy: www.rigb.org/editing-ri-talks...
Subscribe for the latest science videos: bit.ly/RiNewsletter
Product links on this page may be affiliate links which means it won't cost you any extra but we may earn a small commission if you decide to purchase through the link.

[@TheRoyalInstitution]

You might have heard the audio version of this podcast which we recently published to the channel by mistake - but this is the full, video version, we hope you enjoy!

[@Amethyst_Friend]

Wondered about that! Thank you.

[@tizio13]

What a good mistake to have! This was an interesting conversation. This video version is well polished. Hope to see more like it in the future!

[@michaelroseagain]

Wonderful to see behind the scenes of real AI applications. Most AI use seems hidden due to commercial interests or all over the place with images and prompts for very little other than entertainment purposes. Thank you for making this!

[@grahamwilson8843]

Such an amazing interview on a topic that I couldn't have even imagined would ever be possible. It makes me want to study biology. Thanks to The RI, Lisa, and Janet for putting this together!

[@carolspencer6915]

Good afternoon RI and Dame Janet Thornton Super facts in science and AI application explained splendidly. Truly grateful for you and your work. 💜

[@PeteTheProps]

I have enjoyed the podcasts in the past but the video improves them tremendously as I have in the audio only casts wished I could see the body language and enthusiasm your speakers have and in the case of this one the simple but very clarifying examples of the models. Please keep up the excellent output.

[@subliminalvibes]

Wonderful. Thanks for sharing this podcast on KZbin! 👍😎🇦🇺

[@CarlosFernandoCastanedaOlano]

Thank you Royal Institution (RI), greetings from Bimac Research Group at universidad del Cauca, Colombia.

[@corrpearce2363]

Thank you so much for another gift of knowledge

[@JTheoryScience]

such a lovely interview, This was wonderful

[@mowgo9879]

Amazing, thank you!

[@ivyhamilton8879]

So excited for the podcast and love this video version! If I could make a suggestion though, I found the lighting to be quite distracting due to the bright colors. If it could be subdued in future, it'd be perfect for me! Great work!!

[@dphipson]

Fascinating

[@juliecoopermusic-composer6439]

Really interesting podcast thank you Ri 👏🏼

[@juliecoopermusic-composer6439]

Great job producer Lia Hale 😊

[@julsprice8081]

Thank you. This was so interesting - even for a non-scientific golden oldie like me

[@fburton8]

Are there ways to check the predicted protein structure is correct without having to use the techniques (e.g. x-ray crystallography) standardly employed to determine structure? Does the AI _always_ get it right?

[@zephyrandboreas]

Do you mean experimentally? X-ray crystalography is the best method. But other techniques are also very useful and look at other aspects of the structure. It is not just the shape but how the structures 'jiggle' and that is very important to identify rules for how the structure ('shape') relates with function.

[@bonetiredtoo]

A prediction: Alphafold and the people behind it will be the first AI team to win a Nobel prize for chemistry (with the proviso of the three people issue )...

[@charcoal386]

That just tickled my brain a little. I didnt really understand it. Encouraging to see great humans using great tools to help the world.

[@mpen7873]

Nice 👍

[@pauldannelachica2388]

❤❤❤❤❤

[@roqua]

12:18 Interviewer forgot to bring Thornton back to the 2nd problem originally mentioned at the beginning of that explanation. I suspect Thornton forgot she had listed two problems: the first being "folding" and the second being...?

[@roqua]

8:25 being the point at which Thornton mentions that the folding problem has two aspects, the 2nd of which seems to have been left behind.

[@faisalsheikh7846]

Bring demis

[@2CSST2]

I appreciate her expertise but I disagree about her assertion that protein folding still has not been solved. Her argument would be like saying a chess AI that actually played perfect chess would still not have solved chess if it didn't explain to everyone exactly how it finds the perfect move (like in the typical neural net type). Whether it can explain it so you understand it is a different matter than whether it can do it or not. And actually in terms of how proteins fold (or how chess moves are judged), we actually DO know exactly how it works, all it is is the physical laws being followed by the matter involved. So in fact, what she means by "understanding" is really nothing more than finding the right shortcuts of calculation to those physical laws to make their impossible computation something possible to do, in the same way that meteorology doesn't predict the whether by calculating every single particle's trajectory, but instead finds shortcut to that by deriving macroscopic behavior predictions. All of that being said, I don't think calling something solved should have to do with us being provided all the possible calculation shortcuts, they're just that in the end: shortcuts. Making the correct calculation is the thing to solve, not doing it in the most understandable "speaks-to-us" way. We know the actual complete mechanics driving the folding: the physical laws, and we can calculate the end result of those laws.: Solved.

[@TG-lp9vi]

Great video. However the production could be improved. Lighting should come from above to avoid shadows. More soft light too dark.Framing of each speaker should be as such they are facing each other. The editing works well. My two cents

[@Swingingbells]

Nothing wrong with a nice bit of chiaroscuro and some mood lighting. There's nothing at all wrong or substandard about this video's production.

[@trevormatthews7981]

I found some of the shadows distracting particularly the hand shadows over bodies.

[@nickfosterxx]

Personally I enjoyed the lighting, although yes it could be a little distracting, and possibly would work better in a different seating layout.

[@sarcasmo57]

I folded some.

[@tabaks]

Unique, not eunuch.

[@jpf119]

the background music of the intro is unbearable

[@beachboardfan9544]

Whos the cutie interviewing Thornton?