Lex Fridman PodcastDmitry Korkin: Computational Biology of Coronavirus | Lex Fridman Podcast #90
CHAPTERS
- 0:00 – 2:31
Dmitry Korkin’s COVID-19 structural genomics map (why this work matters)
Lex introduces Dmitry Korkin’s background in computational biology and the group’s early-pandemic effort: reconstructing 3D structures of major SARS‑CoV‑2 proteins and mapping interactions with human proteins. The chapter frames the conversation as a tour of how computation accelerates biological understanding and drug/vaccine discovery.
- •Korkin’s research areas: bioinformatics, computational genomics, systems biology
- •Using the SARS‑CoV‑2 genome to model 3D protein structures
- •Predicting virus–human protein interactions (structural genomics map)
- •Rapid, open sharing of models/data to help the global research community
- 2:31 – 6:02
Viruses as “intelligent machines”: simplicity, efficiency, and swarm-like behavior
Korkin describes viruses as non-living but highly optimized machines, whose “intelligence” is expressed through simplicity—doing much with minimal information. Lex pushes on the idea of intelligence, leading to discussion of swarm-like efficiency from micro (virion) to macro (pandemic) scales.
- •Viruses as machines with limited but essential functions
- •“Intelligence” as simplicity and extreme efficiency
- •From nanoscale virions to societal-scale pandemics
- •Speculation about swarm intelligence in viral populations
- 6:02 – 10:48
Natural vs engineered threats: how hard is it to engineer a dangerous virus?
The conversation turns to whether future pandemics are more likely to be natural or engineered. Korkin argues natural emergence is the primary near-term concern, while acknowledging the need for regulation and for studying mechanisms of cross-species jumping to improve prevention.
- •Natural emergence is the dominant historical pattern (flu strains, coronaviruses)
- •Engineering concerns exist, but regulation and oversight are critical
- •Smallpox as a case study in controlled storage and societal safeguards
- •Studying jump mechanisms to enable better (even universal) vaccines
- 10:48 – 22:58
What makes a virus contagious: R0, incubation, asymptomatic spread, and environment
Korkin breaks down contagion as a combination of biological and social factors, including transmission route, surface survival, replication speed, and asymptomatic prevalence. They define R0 and compare smallpox, measles, seasonal flu, SARS‑CoV‑2, and MERS to highlight different risk profiles.
- •R0 explained and why exponential growth matters
- •Factors: transmission route, surface survival, replication dynamics
- •Incubation and asymptomatic shedding as key drivers for SARS‑CoV‑2
- •Trade-off hypothesis: higher pathogenicity often correlates with lower contagiousness
- 22:58 – 27:00
How coronaviruses infect cells: spike–ACE2 binding and hijacking ribosomes
They walk through the basic infection pipeline: attachment to host receptors, membrane fusion, RNA release, and host machinery hijacking. SARS‑CoV‑2 likely binds ACE2 like SARS, possibly with higher binding efficiency, setting the stage for why spike is central to vaccines and therapeutics.
- •Attachment as the first required step of infection
- •Spike protein trimer binding to ACE2 receptor
- •Fusion, RNA release, and dependence on host ribosomes (no viral ribosomes)
- •Coronavirus family context: SARS, MERS, and other strains used in research
- 27:00 – 29:52
Coronavirus parts list: structural proteins, 29 genes, and unknown functions (E protein)
Korkin outlines the virus’s main structural components: spike, membrane, envelope (E), and nucleocapsid proteins, and notes SARS‑CoV‑2 encodes far more proteins than influenza. Despite decades of coronavirus research, key functions—especially for some proteins—remain incompletely understood.
- •Surface proteins: spike, membrane (most abundant), envelope (E)
- •Nucleocapsid protein protects/binds the viral RNA
- •SARS‑CoV‑2 has ~29 proteins vs ~8–9 for influenza
- •Some proteins remain poorly characterized despite long study history
- 29:52 – 40:32
How to infer protein function computationally: similarity, prediction, and experimental validation
Lex asks how scientists determine what a protein does; Korkin explains the computational approach: compare against known proteins, predict structure/function/interactions, then validate experimentally. The chapter emphasizes bioinformatics as a triage and acceleration tool rather than a final authority.
- •Function inference via similarity to previously studied proteins
- •Predictions: structure, function, and interaction hypotheses
- •Bioinformatics streamlines experimental discovery pipelines
- •Hard cases: truly novel proteins with no clear analogs
- 40:32 – 50:07
Protein folding: domains, ab initio limits, Folding@home, and AlphaFold/CASP progress
They discuss why folding is difficult (huge search spaces, multi-domain complexity) and why some proteins (e.g., transmembrane) are especially challenging. Korkin reflects on human intuition in folding games, the role of community datasets like the Protein Data Bank, and machine learning’s rise via CASP and AlphaFold.
- •Proteins often have canonical folds, but conformations can vary
- •Domains as evolutionary/function units; multi-domain proteins add complexity
- •Folding@home and human-in-the-loop folding successes
- •CASP as the folding “Olympics”; AlphaFold’s major leap
- •Protein Data Bank as a foundational open dataset
- 50:07 – 53:27
Open science during the pandemic: preprints, accelerated review, and “knowledge over journals”
The conversation highlights how COVID-19 shifted scientific norms toward faster and more open sharing, with preprints and rapid collaboration. Korkin describes posting early versions quickly (even on a lab website) to get results into researchers’ hands, prioritizing knowledge transfer.
- •Pandemic-driven surge in collaboration and data sharing
- •Preprints and rapid screening/review cycles
- •Publishing as knowledge dissemination rather than journal prestige
- •Practical example: posting an early preprint to a website due to backlog
- 53:27 – 1:07:56
Structural Genomics of SARS‑CoV‑2: from genome sequence to 3D models and protein complexes
Korkin explains how his team began with a raw ~30k nucleotide genome, identified gene boundaries using known coronavirus annotations, and applied homology (template-based) modeling to predict protein structures. They then extended modeling to multimeric complexes (e.g., spike trimer, envelope pentamer) and broader interaction maps.
- •Origin story: class project on another outbreak enabled rapid pivot to SARS‑CoV‑2
- •Genome release as the key starting point; gene boundary identification is straightforward with templates
- •Homology/template-based modeling: sequence similarity implies structural similarity
- •Modeling complexes and multimers: spike trimer, envelope pentamer, other assemblies
- •Interactome concept: predicting virus–virus and virus–host interactions
- 1:07:56 – 1:13:52
Mutations in 3D: conserved drug-binding sites vs variable functional regions
They explore how mutations that look scattered in sequence can cluster in 3D space, often near functional sites. A key finding Korkin emphasizes is that several known small-molecule binding sites appear largely conserved between SARS and SARS‑CoV‑2, supporting drug repurposing hypotheses—pending experimental validation.
- •Sequence differences can cluster spatially in 3D structures
- •Evolution targets some proteins/regions more than others (non-uniform change)
- •Conserved binding sites suggest existing ligands may still work
- •3D structures help identify functional sites and prioritize experiments
- •Predictions must be validated in wet-lab and clinical settings
- 1:13:52 – 1:34:47
From virion structure to intervention: vaccines, nanoparticles, and antiviral drug targets
Korkin describes the virion’s architecture (envelope, spike counts, membrane lattice) and the effort to build whole-particle models informed by microscopy (including ellipsoid vs sphere debate). The chapter bridges structure to interventions: vaccines train antibodies to block spike–ACE2 binding; antivirals block critical viral functions (e.g., polymerase, proteases), with Remdesivir as an example.
- •Virion architecture: lipid envelope with embedded proteins; spike count estimates
- •Microscopy vs idealized models; shape uncertainty and ellipsoid hypothesis
- •Vaccine mechanism: antibodies targeting spike to block ACE2 binding
- •Different vaccine strategies: live-attenuated, stripped/inactivated particles
- •Antivirals: polymerase inhibition (Remdesivir), protease inhibitors, repurposing logic
- •Risk of resistance: mutations in binding sites could reduce drug efficacy
- 1:34:47 – 1:37:17
Is a virus alive? Autonomy, dependence, and the philosophical boundary
Lex challenges the claim that viruses aren’t alive, comparing viral dependence on hosts to human dependence on environments. Korkin anchors his view in autonomy: viruses lack key tools and cannot function independently, even if they exhibit adaptive, evolution-driven behavior.
- •Viruses as non-autonomous entities requiring host machinery
- •Debate over defining “life” via dependence vs autonomy
- •Viruses as a clean illustration of evolution and co-evolution with hosts
- •Discovery of giant viruses complicates origin narratives
- 1:37:17 – 1:50:46
Agent-based epidemiological modeling: “Zombies on a Cruise Ship,” pathogen agents, and intervention insights
Korkin outlines a multi-year agent-based modeling project that began as a cruise-ship outbreak simulation, later applied to real pathogens and even the fictional virus from the film Contagion. The key methodological twist is modeling the pathogen explicitly as an agent, enabling flexible integration of parameters like surface survival, shedding, asymptomatic transmission, and incubation dynamics—crucial for SARS‑CoV‑2.
- •Agent-based models for confined environments (cruise ships, schools, workplaces)
- •Explicit pathogen agent to represent virus behavior alongside host behavior
- •Parameters: transmission routes, fomite survival, shedding profiles, symptomatic vs asymptomatic contagion
- •Models support both forecasting and comparing interventions
- •COVID-19 specific challenge: long asymptomatic/early contagious period changes containment dynamics
- 1:50:46 – 1:55:24
Masks, behavior, and society: from Spanish flu photos to modern compliance
They discuss how uncertain evidence and human behavior shape real-world outcomes, focusing on masks as both self-protection and source control—especially important with asymptomatic spread. Historical reference to 1918 mask usage frames masks as a seriousness signal and collective-action tool, while acknowledging psychological and social friction.
- •Mask benefits: protect wearer and reduce outgoing spread
- •Asymptomatic and fully asymptomatic cases motivate universal masking
- •Uncertainties: airborne/aerosol transmission debates and evolving evidence
- •Historical analogy: Spanish flu mask compliance and enforcement examples
- •Psychological effects: distancing vs signaling care and rebuilding trust
- 1:55:24 – 2:09:01
Russia, scientific collaboration, bobbleheads, and fragility of human life
The conversation closes with Korkin’s personal story—moving from Russia to Canada/US, early ambitions, and a passion for applied mathematics—and a nuanced view of collaboration with Russian bioinformatics groups. Lighthearted discussion of scientist bobbleheads transitions into a reflective ending: viruses reveal human fragility and the need for social cohesion and purposeful science.
- •Korkin’s background: Russia → Canada → US; shift from pilot dreams to math and science
- •Perspective on Russian bioinformatics strength and openness to collaboration
- •Remote collaboration as a silver lining for global science
- •Favorite scientists/bobbleheads: Watson & Crick, Rosalind Franklin, Kolmogorov, Turing, von Neumann, Linus Pauling
- •Final reflection: pandemics highlight fragility, solidarity, and the value of scientific work
