Korkin, Dmitry
▶ Watch on YouTubeCore ideas
Korkin describes viruses as fascinating rather than terrifying — "machines" (not living organisms, in his view, because they're not autonomous and depend entirely on hijacking a host cell's ribosomes to produce proteins) whose apparent intelligence lies in achieving enormous functional efficiency from minimal genetic information, comparing this to swarm intelligence. He expresses more concern about naturally emerging viruses than deliberately engineered ones, noting that new influenza and coronavirus strains emerge regularly through natural evolution, and that engineered pandemic viruses face the same fundamental problem as nuclear weapons: they're inherently difficult to fully control once released, which somewhat self-limits their military usefulness (while affirming controversial gain-of-function research, like published influenza-jump experiments, is still valuable to understand and defend against natural jumps). He highlights the historical contagiousness comparison — SARS-CoV-2's estimated R0 of roughly 1.5–3 versus smallpox's 5–7 and measles' 15+ — and identifies an important, underappreciated dynamic: an apparent inverse relationship between a virus's contagiousness and its lethality (citing H7N9 avian flu's ~30%+ mortality rate but limited human-to-human transmission, and MERS's high lethality without ever becoming pandemic), an open biological question he can't fully explain, alongside long asymptomatic incubation periods and the substantial fraction of entirely asymptomatic-yet-contagious carriers as critical, historically underweighted epidemiological factors.
Korkin explains the practical mechanics of coronavirus infection: the spike protein forms a three-copy "trimer" that binds to the human ACE2 receptor (found in abundance in epithelial cells of the nasal passages, lungs, and other organs), after which the virus fuses membranes and releases its RNA to hijack host ribosomes — since it carries no ribosome of its own. He walks through the virus's key structural proteins (spike, for host attachment; membrane protein, forming the envelope's lattice-like structure; a still poorly understood envelope protein; and a nucleocapsid protein protecting the viral RNA), noting SARS-CoV-2 encodes roughly 29 proteins compared to influenza's 8-9. On his own computational approach, he describes bioinformatics as a discovery-streamlining tool for experimental virologists: since a raw genome sequence alone is uninformative, researchers first identify likely gene boundaries by comparison to known relatives (SARS, MERS, and bat coronaviruses), then use "homology" or template-based modeling — inferring 3D protein structure from sequence similarity to previously solved related protein structures in the open, community-maintained Protein Data Bank — to generate high-confidence structural predictions for proteins that lack templates being a much harder, still largely unsolved "ab initio" problem. He explains protein folding as fundamentally combinatorial and computationally expensive (canonical folds exist per protein domain, but larger multi-domain human proteins and especially transmembrane proteins remain very difficult to model), highlighting the surprising success of the crowdsourced Folding@Home game (where human players sometimes outperformed scientists on specific fold predictions) and DeepMind's AlphaFold (a top performer in the CASP structure-prediction competition) as evidence that decades of accumulated structural data are now enabling genuinely learnable approaches to a historically intractable problem.
Describing his own paper mapping evolutionarily conserved functional regions of SARS-CoV-2 proteins, Korkin recounts its origin in an unrelated 2019 undergraduate bioinformatics class project studying Eastern Equine Encephalitis virus (a mosquito-borne, brain-targeting virus that disrupted daily life in parts of New England but remained a scientifically "neglected disease" due to limited commercial incentive) — a template his team quickly repurposed once the novel coronavirus genome was released in early 2020. Comparing SARS-CoV-2 to its closest relatives (SARS and bat coronaviruses), his team found that sequence differences that appear scattered in the raw 1D genome often cluster together meaningfully once mapped onto 3D protein structure, frequently corresponding to functionally important regions — and, notably, found that binding sites targeted by known small-molecule drugs are largely unmutated relative to SARS, a promising sign that existing antiviral candidates (or structurally similar new ones) could remain effective against the new virus. He discusses remdesivir (originally developed for Ebola/Marburg, working by mimicking an RNA nucleotide to disrupt viral replication) as, in his view, the most promising near-term antiviral candidate, alongside protease inhibitors and repurposed antimalarial compounds, and frames antivirals (rather than vaccines) as the more realistic short-term tool against the ongoing pandemic, given vaccines historically take years to properly develop and validate even under the unusually accelerated ~18-month timelines being pursued for COVID-19.
On broader modeling work, Korkin describes an agent-based epidemiological simulation project (originally built for a lighthearted undergraduate "zombies on a cruise ship" exercise, later applied to norovirus, Ebola, SARS, flu, and even the fictional virus from the film Contagion) whose key innovation was explicitly modeling the pathogen itself as an interacting "agent" alongside human agents — allowing flexible encoding of parameters like surface survival time, transmission mode, and differential viral shedding between symptomatic and asymptomatic hosts, all within a confined, trackable environment (a cruise ship, school, or company) rather than population-scale epidemiology. He argues SARS-CoV-2's unusually long and often fully asymptomatic-yet-contagious incubation period fundamentally changes containment dynamics in ways past pandemic models under-weighted, and — reflecting the era's live scientific debate — argues masks provide meaningful bidirectional protection (protecting others from an unknowingly asymptomatic wearer as much as protecting the wearer), drawing a historical parallel to near-universal mask-wearing during the 1918 Spanish flu. He closes reflecting on his own path from aspiring Soviet-era airline pilot to mathematics and computational biology, praises the continued strength of Russia's mathematical/bioinformatics research community as a source of ongoing collaboration despite geopolitical friction affecting broader scientific exchange, and states that studying how a microscopic entity can threaten all of humanity has deepened his sense of human fragility and interdependence, alongside hope that his and others' scientific work is genuinely useful in response.
Excerpts
On why the virus itself functions like a highly efficient machine rather than a fully "living" organism:
“The divider itself... it's not a living organism, it's a machine — but it is perfected to the way that it essentially has a limited number of functions it needs to do, some necessary functions, as well as the ability to modify itself... the ability to do so much with so little material and information.”
On the small structural mutations that made cross-species jumps and human transmission possible:
“It used to be in bat species, and perhaps because of a couple of small changes in the viral genome, it first had... become capable of jumping from bats to human, and then it became capable of jumping from human to human. It's not even the size of a virus — it's the size of several atoms — and all of a sudden this change has such a major impact.”
On what his structural comparison of SARS-CoV-2 to SARS revealed about drug targets:
“One of the most interesting findings we had was the fact that the viral binding sites on the viral surfaces that get targeted by known small molecules were pretty much not affected at all — and that means that the same small drugs or drug-like compounds can be efficient for the new coronavirus.”