#153 – Dmitry Korkin: Evolution of Proteins, Viruses, Life, and AI

Episode Summary

Executive Summary: Dmitry Korkin discusses protein domains as the real functional/evolutionary building blocks of life, the structural biology of SARS-CoV-2, and how virus evolution informs vaccines, therapeutics, and biosurveillance. The conversation broadens into protein folding, AlphaFold, expert systems, machine learning in bioinformatics, origins of life, alien life, and the ethics of engineering biology and viruses.

Main Topics: Proteins, domains, and modularity (Priority: 5/5): Korkin argues that proteins are best understood as modular assemblies of domains rather than as single globular units. Domains are the key structural, functional, and evolutionary building blocks, while linkers and termini add flexibility and specificity. SARS-CoV-2 structural biology and viral machinery (Priority: 5/5): The discussion explains the spike, envelope, membrane, and nucleocapsid proteins, how the spike trimer binds ACE2, how mutations affect opening dynamics, and why understanding these structures matters for intervention design. Virus evolution, mutation, and therapeutic targets (Priority: 5/5): Korkin emphasizes that viral mutation is the mechanism of host jumping and adaptation, but notes that not all mutations are equally dangerous. He highlights evolutionary stability of the M protein as a promising drug target and discusses vaccine pressure, host shifts, and forecasting future evolution. Protein folding and AlphaFold (Priority: 5/5): The conversation covers the long-standing challenge of protein folding, CASP benchmarking, the role of contact maps, and why AlphaFold is a breakthrough but not the end of the story—especially for multi-domain proteins and protein complexes. Machine learning, expert systems, and bioinformatics (Priority: 4/5): Korkin links current ML successes to earlier expert systems such as DENDRAL and to symbolic machine learning. He argues that successful AI in biology often embeds domain expertise and that bioinformatics is increasingly a search problem informed by prior knowledge. Origins of life, Drake equation, and alien biology (Priority: 3/5): The discussion turns to the likelihood of life emerging on habitable planets, rare-Earth arguments, and whether extraterrestrial life would likely be carbon-based and protein/RNA-based if discovered. Reading, poetry, and intellectual life (Priority: 3/5): Korkin shares formative books and ends by reciting a Russian poem, tying his scientific worldview to literature, memory, and the emotional experience of science.

Key Arguments: Protein domains, not whole proteins, are the cleanest functional and evolutionary units for understanding biology. Protein structure is more conserved than sequence, so evolution often preserves domain-level architecture while changing residues. The SARS-CoV-2 spike is a complex trimer with multiple domains and conformational dynamics; mutation can increase the probability of receptor-binding domains opening. The viral membrane protein (M) may be a promising drug target because it is evolutionarily more stable than spike. Mutations are the natural route by which viruses adapt, jump species, and potentially evade immune pressure; understanding evolution helps forecast future trajectories. AlphaFold is a major breakthrough, but it mainly solves compact domains and does not yet resolve the full complexity of multi-domain proteins and protein complexes. Protein folding remains mechanistically mysterious even if prediction performance has improved, because folding happens quickly and often co-translationally. Expert systems did not simply fail; they evolved into modern AI/ML methods that still incorporate domain knowledge. Nature-inspired computation remains a rich source of algorithms, and future software systems may become more agent-like and evolutionary. The origin of life remains unresolved; probabilities for abiogenesis are highly uncertain, but life may be neither vanishingly rare nor ubiquitous in a simple way. Engineered viruses are possible in principle, but nature is already a highly capable virus designer, and regulation plus experimental constraints matter. Scientific communication and data-sharing during COVID-19 have accelerated dramatically compared with SARS-era timelines.

Data Points: Spike protein copy number per virion: 50 to 90 - Approximate number of spike trimers on a SARS-CoV-2 particle mentioned in the discussion of virion structure. M-dimer copy number per virion: roughly 1000 - Estimate of the number of membrane protein dimers making up most of the outer shell of the virus. E protein copy number per virion: 2 to 3 - Approximate number of envelope pentamer complexes per viral particle. Spike structure resolved by cryo-EM: about two-thirds - Korkin notes cryo-EM reconstructs roughly two-thirds of the spike protein, with the membrane-embedded portion remaining difficult. Protein domains in PSD95: 5 - Example of a large, flexible scaffold protein in synaptic biology used to illustrate multi-domain complexity. Protein domains in spike opening dynamics: 3 receptor-binding domains - The spike trimer has three chains/domains that can open asymmetrically to bind ACE2. CASP competition cadence: every 2 years - Protein structure prediction benchmark used to evaluate methods like AlphaFold. Deep Blue vs. Kasparov: 1997 - Cited as a landmark AI moment when a machine defeated a world chess champion. AlphaFold breakthrough reference: 2018 - The transcript refers to the original AlphaFold’s CASP-era shockwave and later AlphaFold 2 improvements. Nobel Prize for protein folding research: 2013 - Korkin says the first Nobel-related acknowledgement of computational protein folding came in 2013. Joshua Lederberg Nobel Prize age: 33 - Korkin notes Lederberg won the Nobel Prize young and later corresponded with him online. Nature paper on glycine in comet dust: 2018 - Referenced as evidence relevant to prebiotic chemistry and the origin of life.

Pivotal Quotes: "The proteins indeed is the basic unit, biological unit that carries out an important function of the cell." — Dmitry Korkin: Opening explanation of why proteins matter, followed by the argument that domains are the more fundamental unit. "It’s a machine that does a lot of different functions, and many of these functions are sort of nearly perfect, but they are not perfect. And those mutations can make those functions more perfect." — Dmitry Korkin: On viral evolution and why mutations can improve viral efficiency, especially receptor binding and host adaptation. "The protein domains are not only functional building blocks, they are also evolutionary building blocks." — Dmitry Korkin: Core thesis of the conversation on modularity, evolution, and how to think about protein structure.

Implications: The episode frames biology as modular, evolvable, and increasingly computable. For medicine and biotech, that means better targets, better predictions, and more powerful but potentially riskier tools for designing proteins and monitoring pathogens.