Traitwell's Thoughts On The Prospects of Human Genomics, 2023.
A summary of likely extensions and innovations in the human genomics field in the near future. Along with some reflections on what will remain out of reach for now.
Traitwell recently put out its assessment about how it sees genetics over the coming year.
I thought it would share it with you.
General Principles
Just about every human trait is under some degree of genetic control. This has been known for a long time but the penny continues to drop, sector by sector. We now know many of the specific SNPs involved, not just the general fact.
Familiarity breeds comfort, not contempt. What seems outlandish now soon becomes commonplace. Old fears become old-wives’ tales. The envelope expands, driven by human needs. Obstacles are identified and swept aside by the march of innovation.
In the near future entire disciplines will be brought firmly into the fold and will abandon dialectics for ‘genetic lenses’ to understand the world properly and start making some real progress. Tectonic plates are in motion, beneath the surface. After the realignment, everyone will pretend that they knew this stuff all along but were just ... busy with other things. Since human culture co-evolved with genes, you cannot ignore the latter. Blinkers will fall off.
However not all disciplines will follow this trend right away. They will crowd around the pool, looking nervous.
The GWAS Missing Heritability Posse
Though we know that everything is heritable, at about 50% on average per trait, we have only found some of these genes. For a few special traits like cognitive ability we have found a lot of genes, but many remain elusive, with perhaps 17% of variance explained at best. The GWAS posse hunting for the ‘missing heritability’ continues to ride, expand, drive on relentlessly. Better methods will soon be used, combining traits more efficiently to increase statistical power. Information from relatives will not simply be thrown away.
Sample sizes will continue to grow for rare traits like schizophrenia, as data is located and gathered, but must hit a limit because of that trait rarity. However they will saturate inexorably. Beyond that, whole-genome, then long-read sequencing will be used to add structural variants to models, each one of which will be very rare but explain a small piece of the puzzle.
sequencing
Costs continue to fall for sequencing. As long as technology is free to attack a target, it will improve at the rate of Moore’s Law. Chip sequencing was a great start but is good only for not-uncommon variants.
Whole genome cost has fallen dramatically to date, now below $200, and will go down much further. Trait-chasers will include much more of this information in search of rare structural variants. Their challenge is
(i) availability of data (since comparatively few people have been whole-genome sequenced, even fewer using long-read) and (ii) computing resources (the resulting data files run to 100 gig per person. Some but not all of these challenges will be overcome in the near term. Long-term, all will fall.
Demes and Large Effects
Most GWAS studies to-date have concentrated on additive effects. They filter out relatives. This is largely for the same reason that people look under the street-lights for their car keys—easy to see. This does not mean that single genes with large, probably multiplicative, effects do not exist. They are just much harder to find. Probably because, due to the detaching effects of meiosis, they have to evolve inside isolated breeding populations (Demes, or proto-species) of relatives. There they work if they get a foothold by chance and are then driven rapidly to fixation by their high (multiplicative) selective advantage. At that stage they become indistinguishable from additive effects in the population. They are hiding but they are out there, and they will be found sooner rather than later.
Ongoing Evolution
There is a common misperception that humans stopped evolving genetically many years ago. The exact mechanism underlying this surprising stasis is never specified. This is false. Many discoveries will be made in the near future showing, on the contrary, concrete and dramatic examples of very recent human evolution, driven by intensified assortative mating and continued inbreeding.
Deep Ancestry and Ghosts
People want to know as much as possible about their ancestors, even back beyond homo sapiens. Ghost—that is, extinct and unknown populations inferred from shared genes—have emerged from exiting data sets. Earlier hominids like homo erectus, which overlapped with us 100k years ago in Java, may be sequenced soon. The current challenge is to find well-preserved DNA.
Large Language Models Applied to Genetics
Models with enormous numbers of free parameters, such as Large Language Models (LLMs), can comprehend the usage rules latent in a corpus of human language text. Collected gene sequences form, in effect, a corpus of texts. We understand what some very limited parts of the genetic ‘text corpus’ mean, from the perspective of flipping switches on a black box and seeing what happens (natural experiments). Some sub-problems within genetics will be attacked using LLM-like techniques with surprising results. Starting with plant and bacteria genomes. This will require very substantial computing resources—a challenge beyond the reach of all but a few players.
Embryo and Gamete Selection
Public awareness is growing about embryo selection for traits. Fears are being addressed. People become more comfortable with radical new technology once they realize the possibilities. These techniques will also be applied at scale to sperm, not just embryos, in the near future and marketed broadly. A much broader range of traits will soon be filtered for. This will herald a new era of human progress and Promethean vigour.
Epigenomics
Human cells are totipotent, before their epigenetic markers are set to specialize them. (No, these markers are not typically passed on among complex animals, they are explicitly reset with each generation.) Reversing ageing and potentially enabling regrowth and restoration of organs will be achieved by finding the right way to reset the epigenetic markers in organ cells. This is closer within reach than popularly supposed, but will proceed cautiously by picking off the easiest yet most desirable targets to reset.
Human Gene Editing
As successes accumulate in the use of precisely-targeted gene editing technology, social barriers against the practice will continue to fall, spearheaded by the eradication of catastrophic disorders in living individuals. Barriers against making these edits heritable in children will surely fall in the future but 2023 may be too early for that. Until it falls, we condemn those with these syndromes to either a childless future and endless rounds of gene editing in offspring.
Public Policy
Western governments remain tentative when applying genomic knowledge. Since genomes contain a ton of information, that is a ton of information-money left on the table. When the shift comes it will be sudden and comprehensive. Genes and the genomic health of populations are matters of national security. By aggregating data on a large scale, investment decisions can be made more rationally. That data will be gathered in bulk (see below) In the near future, it is China who will emerge as a world-leader in applying this new knowledge.
Aggregate Sequencing
It is a challenge to collect DNA from individuals. They need incentives, and have their own interests. This can be side-stepped through environmental collection of DNA, disregarding individuals and establishing aggregate distributions instead. Sequencing in the aggregate is much cheaper than sequencing individuals, and can be scaled nicely. In the near future this will probably be employed by governments looking to establish trait distributions, without worrying about individuals.