Kellis, Manolis

professor at MIT, head of the MIT Computational Biology Group

Manolis Kellis: Human Genome and Evolutionary Dynamics

Lex Fridman Podcast · Jul 31, 2020

• Manolis Kellis
• human genome
• comparative genomics
• evolutionary signatures
• SARS-CoV-2
• coronavirus evolution
• gene duplication
• digital inheritance
• brain-computer interfaces
• meaning of life

Core ideas

Kellis frames the human genome's deepest beauty as digital inheritance: unlike analog protein concentrations that would dissolve across generations, DNA preserves information discretely, which is why Mendel (not the ancients, who saw only continuous traits like height and skin color) first grasped heredity, and why Ronald Fisher later showed that enough discrete Mendelian factors summed together reproduce the continuous-looking inheritance patterns actually observed. He distinguishes vertical genetic inheritance from humans' uniquely extended horizontal inheritance of ideas and culture, enabled by an unusually long period of neoteny (delayed, malleable brain development) that lets a person raised in any era or culture grow up "completely normal" — a channel he says has been radically accelerated by the internet and, during the pandemic, by scientists' unusually fast, self-correcting, twitter-and-preprint-driven collaboration (contrasted with slower, centralized institutional channels), while cautioning that democratized information also erodes respect for expertise and that "my ignorance is as important as your expertise" is a corrosive attitude.

A large technical throughline concerns his comparative-genomics method of "evolutionary signatures": by aligning genomes across species that diverged from a common ancestor (mammals radiating after the dinosaur-killing asteroid ~65 million years ago), one can infer a region's function purely from its pattern of tolerated mutation, without needing to know the sequence's meaning directly — distinguishing protein-coding regions, RNA structures, and regulatory motifs by the "shape" of their conservation. He describes applying this directly to SARS-CoV-2 by aligning it against 44 related coronavirus strains, concluding that the previously hypothesized ORF10 is not a real protein-coding gene, and discovering a novel overlapping gene hidden in a different reading frame within ORF3a. He explains that outward-facing viral proteins (like the spike protein's S1 receptor-binding domain) evolve far faster than internal machinery because they're what directly contacts a changing environment (a new host species' ACE2 receptor), and that this same fast/slow evolutionary-rate signature, measured from historical cross-species data, predicted that the D614G spike mutation — which had never before persisted across millions of years of mammalian coronavirus evolution — would spread rapidly once the virus adapted to human-to-human transmission (it rose from ~1% to ~90% frequency in observed outbreaks). On the question of engineered origins, he argues the specific genomic lineage involved (and the fact that its recombinant S1 domain shows a naturally occurring mixed ancestry typical of coronavirus evolution generally) makes deliberate design implausible, since no one designing a bioweapon would plausibly start from that particular obscure branch.

Kellis repeatedly frames evolution as having two components — "blind, stupid" random mutation and "super-smart, ruthless" selection — and resists anthropomorphizing viruses as intelligent or malicious, arguing instead that viruses evolve toward not killing their hosts quickly simply because lethal strains fail to spread. He extends this into a broader argument that biological systems achieve resilience through messiness rather than clean engineering: whole-genome duplications (like the one behind baker's yeast, roughly coincident with the evolutionary rise of fruit-bearing plants) create huge redundancy that mostly gets discarded, but the surviving duplicated genes can specialize into new functions — illustrating his claim that "if engineers had designed evolution we would still be perfectly replicating bacteria," since breaking things is often the only path out of a local optimum, a principle he sees mirrored in deep learning's move away from rigid, modular, "compartmentalized" engineering design. He also discusses how immune-system genes get co-opted for non-immune functions (microglia pruning unused synaptic connections, macrophages regulating fat storage), how historical selective pressures leave lasting genomic fingerprints (sickle-cell prevalence tracking historical malaria exposure in Africa; an obesity-linked FTO variant that rose from 2% to over 40% frequency during ice-age scarcity and become a modern liability), and voices measured optimism that humanity's high natural immune diversity makes species-level extinction from pandemics unlikely, while emphasizing that diet and exercise remain more broadly protective than any single "magic bullet" intervention.

On human-machine interfaces, Kellis is optimistic about reading neural output (arguing humans will learn to "speak the machine's language" much as they learned palm-pilot handwriting recognition, or as amputees learn to control prosthetics through trial and error) but skeptical that increasing electrode count alone will let a brain-computer interface bypass the need for actual neuroscience understanding on the input/writing side, given how astronomically under-resolved our understanding of how thoughts are encoded across networks of co-firing neurons remains. He closes on language, meaning, and mortality: drawing on his own etymological approach to language (rooted in his Greek background), he argues that linguistic ambiguity and cross-cultural "misinterpretation" are not flaws but a wellspring of creativity, and that meaning is generated entirely on the receiving end, shaped by each listener's accumulated emotional history. Asked about the meaning of life, he describes organizing a "meaning of life symposium" for his 42nd birthday where 42 colleagues each gave seven-minute answers, concluding that the diversity and mutual consistency of those answers suggested the search for meaning is itself close to the answer — locating his own sense of purpose in intellectual usefulness, gratitude, and watching his children's cognitive and moral development unfold as "raw materials" of humanity being progressively arranged into a fuller person.

Excerpts