Dmitry Korkin: Evolution of Proteins, Viruses, Life, and AI | Lex Fridman Podcast #153

Protein Structure & Complexity

• Proteins are viewed as modular strings of "beads" called protein domains, which serve as the primary functional and evolutionary building blocks rather than the full protein sequence.
• Protein domains remain stable evolutionary units that are shuffled across species, whereas the "linkers" connecting them are flexible regions that facilitate spatial reorganization and interactions.
• The average protein is multi-domain, with complex neurological proteins like PSD95 containing up to five or more domains that act as scaffolds for other proteins.
• Traditional structural determination relied on X-ray crystallography and NMR for smaller, single-domain proteins; cryo-electron microscopy (cryo-EM) now enables the resolution of large molecular complexes like the SARS-CoV-2 spike protein.
• The SARS-CoV-2 spike protein functions as a homotrimer (three identical chains), with recent mutations identified (e.g., by Jeremy Lubin's group) that increase the frequency of multiple receptor-binding domains (RBDs) opening simultaneously, potentially enhancing ACE2 attachment dynamics.

Viral Mechanics & Structural Biology

• The viral envelope is organized by approximately 50–90 spike trimers, 1,000+ M-protein dimers forming a lattice, and a few E-protein pentamers, while the inner nucleocapsid is shielded by N-protein interacting with the M-protein.
• The M-protein (membrane protein) forms a self-organizing lattice structure that is evolutionarily more stable than the spike protein, making it a promising, though less researched, target for therapeutic intervention.
• Roughly one-third of the SARS-CoV-2 spike protein remains structurally unresolved as it is buried within the viral membrane and exhibits instability, though recent low-resolution cryo-EM work suggests the inner capsid (nucleocapsid) may contribute to outer shell stability.

Evolution, Mutations, & Future Risks

• Viral jumps between species (e.g., human to dog) drive mutations; while these can be neutral, they carry the risk of acquiring traits that become damaging or resistant upon reversion to the human host.
• There is no strong evidence currently suggesting vaccine rollout is driving aggressive mutation rates, though scientists monitor the potential "arms race" between vaccination speed and viral resistance.
• Machine learning is being applied to identify molecular determinants of pathogenicity to forecast which mutations might increase a virus's ability to damage human systems or jump species.
• Dmitry Korkin expresses a preference that natural pandemics remain a greater long-term threat than engineered ones, citing the difficulty of targeting a virus and the lack of sufficient motivation for weaponization, while advocating for government transparency to mitigate secrecy risks.

AlphaFold & AI in Biology

• AlphaFold2's achievement in solving protein folding is considered a top-tier breakthrough in AI history, comparable to Deep Blue or the original ImageNet moment, due to its ability to predict structures at near-experimental accuracy for compact domains.
• The "solving" of protein folding is defined by achieving experimental-level performance on the CASP (Critical Assessment of Protein Structure) competition, though predicting the folding of highly multi-domain proteins remains an unsolved challenge.
• AlphaFold's methodology relies heavily on evolutionary data (multiple sequence alignments) to infer "contact maps" (distance matrices) between amino acids, leveraging the fact that protein structure is more conserved than sequence.
• While protein design (engineering new structures/functions) is a primary application, Korkin notes that experimental validation remains the bottleneck, and regulatory frameworks (like "gain-of-function" moratoria) are necessary to prevent dual-use risks.

Origins of Life & Extraterrestrial Perspectives

• Korkin estimates the probability of life emerging on a habitable planet at less than 1%, aligning closer to the "Rare Earth Hypothesis" which posits Earth's specific conditions are unusually favorable for complex life.
• The discovery of glycine in the comet Churyumov-Gerasimenko suggests the building blocks of proteins may be extraterrestrial, yet the universality of carbon-based life remains a key variable in the Drake Equation.
• Finding carbon-based life on Mars or Europa would be a definitive turning point for science, proving life is not unique to Earth, whereas finding radically different biochemistry would suggest life is ubiquitous and inevitable.

Historical Context & Expert Systems

• Joshua Lederberg pioneered "Dendral" in the 1960s, an expert system designed to deduce molecular structures from mass spectrometry data, effectively founding the fields of cheminformatics and bioinformatics.
• Expert systems did not fail due to incompetence but rather because they became too successful, evolving into modern machine learning tools that embed domain knowledge; AlphaFold is the current evolution of this "knowledge engineering" approach.
• The history of AI includes milestones like Deep Blue (search-based intelligence) and Deep Rembrandt (artistic replication), though AI has yet to achieve a commercial "breakout" in music or cinema comparable to human-created content.

Personal & Future Outlook

• Korkin cites The Master and Margarita (Bulgakov), Cancer Ward (Solzhenitsyn), and The Computer and the Brain (von Neumann) as transformative works, with an honorable mention for Lab Girl by Hope Jahren for its personal narrative on scientific life.
• His primary resolutions for 2021 include prioritizing family time and restoring international travel, specifically to attend the School for Molecular and Theoretical Biology in Europe.
• Korkin recites the poem "Sorceress" (Vedma) by Dmitriy Kemerfeld, drawing parallels between the poem's magical winter imagery and the atmosphere of The Master and Margarita.