ResearchPublished Top storyAll news from that day
DeepMind's Kohli says replicating structures does not fully explain protein dynamics
On a Latent Space panel with Biohub's Sal Candido, published 9 October, Google DeepMind's Pushmeet Kohli said replicating deposited structures does not mean protein dynamics are understood.
Check our sources · 14 facts from 3 sources
Video: Latent Space on YouTube
Key points
- Kohli said there have been advances at a conceptual level, but replicating a PDB structure "does not mean that we have understood all of protein dynamics" (17:44).
- Recalling discussions with John Jumper, Kohli said the true ground state that proteins take, and the actual distribution of structures they take, are not known (17:17).
- Kohli asked: "let's not stop the funding of protein structure prediction and protein dynamics", because in his words the field is just getting started (18:15).
What happened
Pushmeet Kohli, listed by Google as Vice President at Google DeepMind and Google Cloud's Chief Scientist, spoke on a panel with Sal Candido of Biohub, moderated by Brandon Anderson. The Latent Space podcast published the video on its YouTube channel on 9 October 2026.
The moderator asked whether another big leap in protein structure prediction is coming. Kohli said that when people say the protein folding problem has been solved, there have been some advances at a conceptual level. He said proteins are extremely complex and disordered, and that their shape can change with context.
Recalling discussions with John Jumper, he said that nobody knows the actual ground state of proteins or their actual distribution of structures. He described the task as one where someone has deposited a structure in the PDB and the aim is to replicate it, and said that does not mean protein dynamics are understood.
He asked that funding for protein structure prediction and protein dynamics not stop.
On whether models need to be understood to be trusted, Kohli said AlphaFold is not perfect and that the calibration of the uncertainty measure was extremely important. He said that when DeepMind released AlphaFold 2 and made the weights available, people found it was a good predictor of disordered proteins, which the team had not discovered before launch.
Asked how long until AI results reach the clinic, Kohli said AI is already used in every part of the drug discovery process. He said the larger acceleration will be unlocked only with a better understanding of biological models.
What it means for you
Our viewThis is Kohli's view on a panel, not a published finding. He separates reproducing a structure someone has deposited in the PDB from knowing how proteins move and change shape, and he says the larger acceleration in drug discovery depends on better biological models.
He also says the calibration of AlphaFold's uncertainty measure was extremely important. If you rely on predicted structures in drug or materials work, check whether your tool reports a confidence score for each prediction, and ask the vendor how that score was calibrated and whether it covers disordered proteins.
It adds no new facts.
Your reaction
We count reactions per story and day, never who reacted. The counts help us choose what goes in the monthly issue. If you are signed in, your own page shows yours too.
We count reactions per story and day, never who reacted. They help choose what goes in the monthly issue.
Check our sources
Every sentence above is checked against these 3 sources.
1 Pushmeet Kohli
Open the source archived copy-
Google's author page for Pushmeet Kohli gives his title as "Chief Scientist, Google Cloud and Vice President, Google DeepMind".
cite
2 Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub (episode page)
Open the source archived copy-
The Latent Space episode page says the video is a special panel with Google DeepMind's Pushmeet Kohli and Biohub's Sal Candido, moderated by Brandon Anderson.
cite
3 https://www.youtube.com/watch?v=NufiHZfMwaw
Open the source archived copy-
The video "Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub" was published on the Latent Space YouTube channel on 9 October 2026 (Pacific time).
citeWatch page metadata: title: Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub; channel: Latent Space; publishDate: 2026-10-09T17:30:41-07:00
-
The moderator asked, addressing Kohli by his first name: "is there another big leap being made here?" (15:56).
cite(15:53) curious uh push me what do you think the (15:56) the like is there another big leap being (15:59) made here?
-
Kohli said: "proteins are extremely complex uh they are disordered uh they their shape might change depending on the context" (16:57).
cite(16:57) proteins are extremely complex uh they (17:00) are disordered uh they their shape might (17:04) change depending on the context
-
Kohli said, recalling discussions with John Jumper: "we don't know what is the actual true ground state that proteins take and what is the actual distribution of structure that proteins take" (17:17).
cite(17:11) John jumper and I basically (17:12) used to sort of discuss this ... (17:17) don't know what is the actual true (17:19) ground state that proteins take and what (17:22) is the actual distribution of structure (17:25) that proteins take.
-
Kohli said: "What we are trying to do is basically someone got a structure, deposited it in the PDB and we are trying to replicate and get the same structure" (17:25), and "that does not mean that we have understood all of protein dynamics" (17:44).
cite(17:25) What we are trying (17:28) to do is basically someone got a (17:31) structure, deposited it in the PDB and (17:35) we are trying to replicate and get the (17:37) same structure. ... (17:44) but that does not mean that we have (17:46) understood all of protein dynamics.
-
Kohli said: "let's not stop the funding of protein structure prediction and protein dynamics because we are just getting started" (18:15).
cite(18:15) to celebrate but uh let's not stop the (18:18) funding (18:19) of protein structure prediction and (18:21) protein dynamics because we are just (18:23) getting started.
-
Kohli said: "When we launched Alpha Volt 2 and made the the weights available, people found out that it was a great disordered protein predictor" (26:22), the automatic captions' rendering of AlphaFold 2, and: "we did not sort of discover it before launching" (26:18).
cite(26:18) generalize and we did not sort of (26:21) discover it before launching. When we (26:22) launched Alpha Volt 2 and made the the (26:25) weights available, people found out that (26:29) it was a great disordered protein (26:32) predictor.
-
Kohli said: "AI is being used today already in every part of the drug discovery process" (29:02), after the moderator asked how long until AI results reach the clinic (28:34).
cite(28:46) until we start seeing uh AI results in (28:50) the clinic um push me you want to start ... (29:02) it's it's not a u AI is being used today (29:07) already in every part of the drug (29:11) discovery process
-
Kohli said: "the actual sort of larger acceleration that will be unlocked only with a better understanding of the biolog biological models that this effort is trying to sort of create" (30:14), as the automatic captions give it.
cite(30:14) time. But like the actual sort of larger (30:17) acceleration that will be unlocked only (30:20) with a better understanding of the (30:22) biolog biological models that this (30:26) effort is trying to sort of create.
-
Kohli said of AlphaFold: "is not perfect right it's not perfect" (24:59).
cite(24:56) So alpha (24:59) fold actually is not perfect right it's (25:02) not perfect it's 90 GDT alpha fold 2
-
Kohli, asked about the protein folding problem (16:22), said: "when people sort of say the protein folding problem has been solved like at at a conceptual level yes there might be sort of yes there has been there have been some some advances" (16:25), as the automatic captions give it.
cite(16:12) push me ... (16:14) Yeah, I think science how it sort of (16:17) operates is basically by isolating (16:20) something and then making progress step (16:22) by step. So when people sort of say the (16:25) protein folding problem has been solved (16:27) like at at a conceptual level yes there (16:30) might be sort of yes there has been (16:33) there have been some some advances
-
Kohli, on the confidence score of AlphaFold 2, said that without calibration you "will be working on it for the next one year and finding out it was completely wrong. The calibration of the uncertainty measure was extremely important" (25:32).
cite(25:15) but the PD the PLDDT score was (25:18) completely unccalibrated ... (25:26) tell you here's very very confident and (25:29) you will be working on it for the next (25:31) one year and finding out it was (25:32) completely wrong. The calibration of the (25:35) uncertainty measure was extremely (25:38) important.
Every sentence above is checked against these 3 sources. Each fact has its own address you can cite.
Source 1 · Google
Pushmeet Kohli
Google's author page for Pushmeet Kohli gives his title as "Chief Scientist, Google Cloud and Vice President, Google DeepMind".
Source 2 · Latent Space · 10 Oct 2026
Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub (episode page)
The Latent Space episode page says the video is a special panel with Google DeepMind's Pushmeet Kohli and Biohub's Sal Candido, moderated by Brandon Anderson.
Source 3 · 9 Oct 2026
https://www.youtube.com/watch?v=NufiHZfMwaw
The video "Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub" was published on the Latent Space YouTube channel on 9 October 2026 (Pacific time).
Watch page metadata: title: Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub; channel: Latent Space; publishDate: 2026-10-09T17:30:41-07:00
The moderator asked, addressing Kohli by his first name: "is there another big leap being made here?" (15:56).
(15:53) curious uh push me what do you think the (15:56) the like is there another big leap being (15:59) made here?
Kohli said: "proteins are extremely complex uh they are disordered uh they their shape might change depending on the context" (16:57).
(16:57) proteins are extremely complex uh they (17:00) are disordered uh they their shape might (17:04) change depending on the context
Show all 12 factsShow fewer facts
Kohli said, recalling discussions with John Jumper: "we don't know what is the actual true ground state that proteins take and what is the actual distribution of structure that proteins take" (17:17).
(17:11) John jumper and I basically (17:12) used to sort of discuss this ... (17:17) don't know what is the actual true (17:19) ground state that proteins take and what (17:22) is the actual distribution of structure (17:25) that proteins take.
Kohli said: "What we are trying to do is basically someone got a structure, deposited it in the PDB and we are trying to replicate and get the same structure" (17:25), and "that does not mean that we have understood all of protein dynamics" (17:44).
(17:25) What we are trying (17:28) to do is basically someone got a (17:31) structure, deposited it in the PDB and (17:35) we are trying to replicate and get the (17:37) same structure. ... (17:44) but that does not mean that we have (17:46) understood all of protein dynamics.
Kohli said: "let's not stop the funding of protein structure prediction and protein dynamics because we are just getting started" (18:15).
(18:15) to celebrate but uh let's not stop the (18:18) funding (18:19) of protein structure prediction and (18:21) protein dynamics because we are just (18:23) getting started.
Kohli said: "When we launched Alpha Volt 2 and made the the weights available, people found out that it was a great disordered protein predictor" (26:22), the automatic captions' rendering of AlphaFold 2, and: "we did not sort of discover it before launching" (26:18).
(26:18) generalize and we did not sort of (26:21) discover it before launching. When we (26:22) launched Alpha Volt 2 and made the the (26:25) weights available, people found out that (26:29) it was a great disordered protein (26:32) predictor.
Kohli said: "AI is being used today already in every part of the drug discovery process" (29:02), after the moderator asked how long until AI results reach the clinic (28:34).
(28:46) until we start seeing uh AI results in (28:50) the clinic um push me you want to start ... (29:02) it's it's not a u AI is being used today (29:07) already in every part of the drug (29:11) discovery process
Kohli said: "the actual sort of larger acceleration that will be unlocked only with a better understanding of the biolog biological models that this effort is trying to sort of create" (30:14), as the automatic captions give it.
(30:14) time. But like the actual sort of larger (30:17) acceleration that will be unlocked only (30:20) with a better understanding of the (30:22) biolog biological models that this (30:26) effort is trying to sort of create.
Kohli said of AlphaFold: "is not perfect right it's not perfect" (24:59).
(24:56) So alpha (24:59) fold actually is not perfect right it's (25:02) not perfect it's 90 GDT alpha fold 2
Kohli, asked about the protein folding problem (16:22), said: "when people sort of say the protein folding problem has been solved like at at a conceptual level yes there might be sort of yes there has been there have been some some advances" (16:25), as the automatic captions give it.
(16:12) push me ... (16:14) Yeah, I think science how it sort of (16:17) operates is basically by isolating (16:20) something and then making progress step (16:22) by step. So when people sort of say the (16:25) protein folding problem has been solved (16:27) like at at a conceptual level yes there (16:30) might be sort of yes there has been (16:33) there have been some some advances
Kohli, on the confidence score of AlphaFold 2, said that without calibration you "will be working on it for the next one year and finding out it was completely wrong. The calibration of the uncertainty measure was extremely important" (25:32).
(25:15) but the PD the PLDDT score was (25:18) completely unccalibrated ... (25:26) tell you here's very very confident and (25:29) you will be working on it for the next (25:31) one year and finding out it was (25:32) completely wrong. The calibration of the (25:35) uncertainty measure was extremely (25:38) important.
Topics
The morning email
On the mornings we publish: the three top stories and up to four short ones. Free.
We email you a link to confirm. An issue may include one sponsor, always labelled Sponsored · Advertisement. Our emails count opens and clicks, not who made them. Unsubscribe in one click. What we keep