This is a list of projects I’d like to assist.
This is a note aimed mainly at scientists, but also funders
(philanthropy, government, commercial), and separately at other experts
(law, policy, regulation, ethics, media). I’d like to assist with
various scientific projects related to advanced reproductive technology,
including by trying to direct funding to those projects.
Relevant scientific areas:
- Reproductive epigenetics. (Epigenetic sequencing
and editing, epigenetics of reproduction and the germline, stem cell
culturing, gonadal culturing, stem cell reprogramming, gametogenesis,
creating gametogonia.)
- Chromosome engineering. (Targeted crossover,
targeted elimination, targeted missegregation, chromosome transfer,
microinjection, nuclear transfer, manipulating and sorting chromosomes,
physics of individual chromosomes.)
- Microfluidics. (Cell lysis, microwells, droplet
creation / transportation / sorting / merging, PDMS design and
manufacturing.)
- Cell engineering. (Stem cell culturing, DNA damage
and repair, CRISPR-Cas9 and transposases and other gene editing
systems.)
- Statistical genetics.
I’d like to assist with projects that have a good chance at
meaningfully accelerating the creation of the science, technology, and
social conditions needed for the socially beneficial deployment of
strong reprogenetics, and in particular the ability for parents to vector
the genome of their future child by several standard deviations on
cognitive ability. This note lists a wide range of projects that I’d
like to assist, whether by advising, collaborating with, or directing
funding to such projects.
In general, it’s hard for an outsider to accelerate a field like
reprogenetics. That’s because it mostly consists of various fields of
science and technology that are already being worked on using
significant resources: polygenic prediction, stem cell bio, in vitro
gametogenesis, gene editing, epigenetic editing, chromosome transfer,
DNA sequencing, etc. However, I think there are many opportunities to
accelerate the field, given the right intentions and funding. Many key
projects are underfunded because government funding has restrictions and
recent cuts; venture funding is skittish due to concerns about
regulation or due to the science being not ready for translation to
industry; philanthropic funding has PR concerns; and academia and
industry may also neglect public goods that confer insufficient profit
or academic prestige, such as datasets, cell lines, oversight, and
public discourse.
I want to help fill those gaps directly. I also want to assist with
projects that work to improve the social and regulatory landscape, in
ways that would help fill those gaps in general. I’m open to being
convinced that my picture about how to accelerate that is wrong or
incomplete.
My main analysis of the technical problem of strong reprogenetics is
here: https://berkeleygenomics.org/articles/Methods_for_strong_human_germline_engineering.html.
I’m interested in critiques of the main conclusions there, especially if
they’d change what projects seem good to support.
In order to help those gaps be filled, here’s a list of some specific
projects. I’d like to assist with these projects; and, to anyone who
wants to accelerate reprogenetics, I’d suggest that these are good
projects to assist.
(Note that the following projects are somewhat selected for being in
biotechnology’s “valley of death”. There’s lots of good relevant
research in academia, and there are several promising startups worth
investing in. More ambitious ways to accelerate the field would become
feasible with more funding.)
- There are many potential perils of reprogenetics. These perils could
be analyzed (for severity, likelihood, and prevention and alleviation
methods). See “Potential
perils of germline genomic engineering”.
- What governance structures should be used for research groups and
vendors developing reprogenetics technologies?
- Talking to stakeholders (the public, advocacy groups, etc.) to
understand opinions around reprogenetics, e.g. fears, hopes, questions,
confusions, requests, etc.
- How should society develop and deploy reprogenetics technologies?
- What regulations and social attitudes should be used?
- What process should society use to decide this?
- How to ensure that reprogenetics…
- …doesn’t get misused?
- …doesn’t harm recipients?
- …isn’t marketed via fraud?
- …is widely accessible?
- …is broadly beneficial for society?
- Doing fundraising, whether from VC, large private capital,
philanthropy, or government sources. Also, mapping the views of these
funders—what preconditions would have to be met to get much more funding
for key science and technology research.
- Doing advocacy for reprogenetics, broadly construed, e.g. producing
media that explains and discusses reprogenetics to help society process
these possibilities.
- Gathering together people who have expertise and legitimacy to map
out plausible futures of a given technology, discuss safety and ethics,
make open letters, roundtables, reports, etc.
- How does reprogenetics affect existential risk from AGI? (See my
essays here; they leave a lot of work to be done: https://tsvibt.blogspot.com/2025/11/hia-and-x-risk-part-1-why-it-helps.html,
https://www.lesswrong.com/posts/K4K6ikQtHxcG49Tcn/hia-and-x-risk-part-2-why-it-hurts)
- Doing social organizing and momentum-building. E.g. conferences,
forums, etc.
I’m not sure about many examples here, but in general if there are
very good proposals for activities that would support the field as a
whole, I’m interested. Some examples:
- Creating a primate research center focused on novel reproductive
technology. If the OHSU primate center shuts down, this will be a
crucial gap to fill in order to test novel reprogenetic
technologies.
- It would be valuable for long-term health and trust in the
reprogenetics sector to have one or more legitimate, expert,
industry-independent organizations to evaluate the relevance and truth
of claims made by reprogenetics vendors. E.g. a third-party validator of
claims about polygenic scores (predictive power, causal validity,
accurate presentation), run by academic experts in polygenic prediction
and possessing a held-out dataset, would help prevent and punish fraud,
build trust by parents, and give regulators an expert reference
point.
- Creating academic and/or open-source versions of technologies might
help drive down prices and improve trust (by making the technology more
transparent and validatable).
- Crafting regulation could be important for paving the way for more
funding and talent in the space. For example, creating anti-trust rules
could help ensure a competitive market and accessibility; fraud rules
could protect parents and build trust; safety standards for editing,
IVG, and so on could assure venture capital that the technologies have a
viable pathway to testing and approval, as well as point scientists to
what has to be done; etc. I’m interested in good ways to approach this,
e.g. convening expert panels, drafting ideas, talking with policymakers,
etc.
Strong reprogenetics probably requires actively correcting the
epigenomic state of non-gamete/zygote cells to be competent as
gametes/zygotes. See https://berkeleygenomics.org/articles/Methods_for_strong_human_germline_engineering.html#reproductive-gv-and-epigenomic-correctness-ec.
There are several approaches to solving or routing around this
problem. Please see here for details: https://berkeleygenomics.org/articles/Methods_for_strong_human_germline_engineering.html#methods-to-handle-epigenomic-correctness.
These approaches point at several research projects.
E.g.:
- Epigenetic editing. (High efficiency, high precision; able to apply
to all necessary types of epigenetic marks (DNA methylation, histone
methylation, other histone modifications))
- In vitro spermatogenesis and oogenesis via
reconstituted organs, xeno organs, organoids, or signaling methods.
Ideally, efficient and fast induction of reprogramming, especially
sex-specific imprinting, probably via molecular signaling methods.
- Spermatogonia extraction, culturing, and retransplantation, while
preserving the ability to differentiate into epigenomically normal
sperm.
I’d also like to assist projects to understand the necessary
epigenomic states involved, as well as the developmental trajectories
involved. In particular, more complete epigenetic cell atlases of human
and primate reproductive tissues during critical germline development
periods would be helpful.
If there’s theoretical work that could plausibly bring great clarity
to the nature of gene regulatory networks, how they are dynamically
navigated and maintained during natural cell differentiation /
reprogramming, and how to efficiently artificially manipulate them, I
might be interested.
In order to carry out a genomic vectoring protocol like iterated
recombinant selection, some way of causing cells to undergo meiosis
would be needed. More generally, causing chromosomes to recombine in
large segments is useful for increasing the peak of available
chromosomes’s scores on PGSes.
I’m interested in projects that make recombination very efficient.
This means developing a protocol that makes stem cells produce daughter
cells with chromosomes produced by crossovers from the parent
chromosomes. Such a protocol should also avoid introducing many de novo
mutations. Inducing meiosis is one potential principal method; other
methods could include random recombination, targeted recombination
(e.g. via DSBs at the same site on two homologous chromosomes and NHEJ
with swapped ends), or hyperrecombination (inducing many crossovers per
chromosome within one meiosis).
See https://berkeleygenomics.org/articles/Methods_for_strong_human_germline_engineering.html#method-chromosome-selection
and see https://berkeleygenomics.org/articles/Chromosome_identification_methods.html.
In particular, I’m interested in methods that have a good chance at
one or more of:
- Assembling several chromosomes from several different cells (as
opposed to just replacing one chromosome).
- Being fast and inexpensive, and preserving genomic integrity (as
opposed to culturing cells for months on end).
- Operating on sperm chromosomes and preserving epigenomic integrity.
(This is not necessary, it’s just a bonus to maybe bypass the need for
epigenomic correction.)
My belief is that MMCT
is unlikely to meet these criteria, but I’m interested in having my mind
changed. Similarly for whole
cell fusion plus random ploidy reduction.
(Whole cell fusion plus targeted chromosome elimination is an
interesting possible alternative. Likewise targeted missegregation.
Eliminating single target chromosomes has been demonstrated in the
context of MMCT; see e.g. Petris et al. (2025). If
there’s a plausible method to eliminate an entire haploid or diploid
chromosome set from a diploid or tetraploid cell, especially in a
targeted manner, without killing the cell, that would be quite
interesting.)
More concretely, I’m interested in hearing from projects working on
microfluidics for analysis of large subcellular biological particles,
and of course in particular, chromosomes. For example, it would be great
to see projects that replicate / refine / develop the following
tools:
- Key challenge: Storing and transporting chromosomes through
microfluidic channels, whether naked or in water/oil or
water/oil/water droplets, without the chromosomes being damaged.
- Damage would include large scale breakage, point mutations,
deproteination, denaturation, or (ideally) epigenetic state
changes.
- Sources of damage would include mechanical forces (fluid shear, wall
sticking), natural deproteination, deamination and other degradation,
PDMS leaching, etc.
- Remedies might include protective membranes; cooling / freezing;
storing (during complementary identification); optimized buffer; special
buffer additives (such as repair molecules and other maintenance
mechanisms as in cells).
- See for example Takahashi et al. (2018).
- Extracting chromosomes from cells while maintaining origin-index
information, e.g. Lam et al. (2024).
- Isolating single chromosomes in droplets, selecting some droplets,
and merging droplets, e.g. Babahosseini et al. (2021). In
particular, a demonstration of ensembling a full euploid haploid genome
into one droplet.
- Single-cell RNA sequencing for sperm, e.g. Bhutani et al. (2021). (Cf. “Chromosome
identification methods—Sequencing post-meiotic RNA”.)
- Efficient sorting of chromosomes from other cellular debris,
e.g. using dielectrophoresis.
- An end-to-end demonstration of complementary chromosome
identification, as described in “Chromosome
identification methods—Chromosome-wise complementary
identification”, and subsequent refinements for efficiency and
keeping chromosomes intact. Concretely, this involves isolating
chromosomes from a single cell and transiting to a sequencing area with
low dropout / wall sticking. See for example Fan et al. (2011).
- Methods to apply optical tweezers at scale (e.g. 64 operations in
parallel). Optical tweezers are interesting as a way of manipulating
chromosomes, but they seem to suffer from a high cost of operation and
low throughput.
The most interesting versions of these projects would use human
condensed chromosomes, and would track whether the chromosomes remain
intact. More generally, I’m interested in projects that refine these
tools to be more effective, more efficient, more accessible, more
reusable, less expensive, etc.
More projects I’d like to fund:
- Experiments that inject a full disaggregated haploid genome (i.e. 23
chromosomes, but not all stuck together as in a natural sperm nucleus)
into an oocyte, activate the oocyte, and see what happens and whether a
male pronucleus forms around all those chromosomes. Along the lines of
Kuretake et al. (1996), but with disaggregated
chromosomes.
- A method for directly non-destructively measuring the number and
allelic content of an isolated condensed human chromosome. I suspect
this isn’t really feasible, but for example, Raman spectroscopy might
work in the vein of Ojeda et al. (2006);
cf. “Chromosome
identification methods—Raman spectroscopy”.
- Investigate ways to isolate a single chromosome from a sperm nucleus
with minimal disruption, e.g. no breaking and minimal deproteination.
This might use some mechanical methods and/or some way of breaking down
specifically protamine-protamine disulfide bonds without removing
histones or modifying chemical modifications to DNA or histones. (Note:
It would be nice, but not necessary, to preserve the epigenetic state of
sperm chromosomes. The other reason to operate on sperm chromosomes is
simply that those are the chromosomes that have undergone crossover,
providing variation to select on. One could create a stem cell line from
selected sperm chromosomes, and then derive an epigenomically corrected
gamete from that stem cell line.)
- Tests of chromosome selection for strong genomic vectoring to
greatly increase polygenic traits in agricultural species (where the
epigenomic correctness problem can be largely ignored). Any chromosome
selection method could be used, including ones that ignore issues of
epigenomics and issues of genomic degradation from de novo mutation. For
example, whole
cell fusion interleaved with ploidy reduction is an inefficient but
perhaps feasible method for chromosome selection that could be tried in,
say, cows today.
Most strong reprogenetic methods would require growing cells in
culture for multiple months. That includes iterated meiotic selection,
iterated CRISPR editing, and many chromosome selection methods.
When cells divide, their DNA gets damaged. The rate is high:
something like, on average, at least 1 single-base-pair substitution per
division (probably significantly more, like 3+), even under somewhat
optimized conditions. Further, there’s some chance of
indels, occasional copy number variation, and possible mitochondrial
mutations. Worse, some mutations would be positively selected for in
vitro. Even further, important operations such as inducing pluripotency
in a somatic cell introduce base substitutions at a higher rate than
mitosis; and meiosis might introduce small rearrangements (though maybe
rarely, e.g. in <10% of divisions perhaps).
In the context of trying to produce genomically vectored gametes,
this is potentially a major issue. Culturing for several months might
introduce de novo mutations (of an unknown nature and impact) at a very
high rate compared to natural mutation. For example, a 20 year old man’s
sperm might have a couple dozen mutations; a 70 year old man’s sperm
might have well over a hundred de novo mutations; a stem cell population
cultured for several months might have hundreds of de novo mutations
(especially cancerous ones).
Questions / projects:
- Are there ways to culture cells with much lower damage rates / much
better repair?
- How bad is this actually, given that natural reproduction does
involve dozens of de novos per generation, presumably usually mostly
inconsequential? (Note that children of very old men in fact have
significantly elevated risks of various issues, including mental
illness, probably partly due to general de novos, partly due to
selection effects for more proliferative spermatogonia, and partly due
to other things like epigenetic sperm health.)
- What sort of gene editing technology could possibly keep up with
this rate of mutation in practice? Is it possible to keep a population
of stem cells alive and within, say, 50 Hamming distance of the starting
genome indefinitely?
- What methods could produce strong genomic vectoring without running
up against this problem? (My very speculative answer is
microfluidics-based chromosome selection; what are other ideas?) For
example, can iterated meiotic selection have large (say, 6 raw SDs)
selection effects with only a couple rounds (at most 2 months of
culturing, say)?
To a large extent, perhaps strangely, I don’t view it as crucial to
do more studies on genetics. We probably know enough about the genetics
of intelligence to greatly increase the expected intelligence of a given
future child. That said, there are a couple projects that I could be
interested in as high-ish priorities:
- Third-party PGS claim validators, as mentioned
above.
- Tail studies.
- A key claim supporting a hope for strong human intelligence
amplification via strong reprogenetics is that because we know very many
variants causal for IQ, amounting to 10s of SDs in the linear model, we
could get at least up to 6 SDs above the mean. This claim does not at
all require that the effects of these variants stay linear. It
just requires that they continue to add up substantially; in other
words, the returns don’t diminish super sharply above, say, 4 SDs.
- A further claim is that such reproductive genomic vectoring would
probably be safe because (1) the variants in question are common
variants, (2) the phenotype in question is within the human envelope,
(3) you’d also genomically vector for general health and also
brain-protective variants, and (4) developmental canalization for
well-formed brains should be reasonably effective (other phenotypes seem
to be ok for somewhat-non-Gaussian trait values).
- Both these claims are however not empirically validated. Studies
that evaluate these claims could be valuable for understanding
feasibility and safety of strong human intelligence amplification via
reprogenetics.
- See also: https://berkeleygenomics.org/articles/Some_reprogenetics_related_projects_you_could_help_with.html#:~:text=if%20you%E2%80%99re%20interested.-,Can%20genomic%20vectoring%20have%20large%20effects%3F,-Many%20scientists%20say
- Cross-ancestry PGSes. It’s important, for wide
accessibility and equality, that there be good polygenic predictors for
anyone who wants reprogenetics. By my lay understanding, currently,
PGSes don’t fully generalize across ancestry groups. I’m interested in
cost-efficient ways to improve this situation (e.g. via better
causal-variant identification) that others aren’t pursuing already.
- Unknown pleiotropy. My understanding is that, so
far, the empirically observed pleiotropies between traits are pretty
small overall, and furthermore generally tend to be of mixed sign
(e.g. many disease risks are slightly positively correlated with each
other, and intelligence is slightly anti-correlated with many mental
illnesses). Further, the known associated genes largely don’t overlap,
suggesting that in general even very polygenic traits tend to be
disjoint (as might be expected from simple models of random sparse
vectors). However, in principle these traits may have just not been
compared to some other less-studied traits; so in principle it could be
that if you substantially intervene on genes associated with some trait,
you also unknowingly significantly increase some other trait by
surprise, which could harm the wellbeing of the future child. How
plausible is this, given what we know? Are there traits that are
important and could plausibly be affected this way, but that we wouldn’t
notice as being correlated with traits targeted with reprogenetics? What
sort of further investigations could inform these questions?
- Meaning and genetics of non-abusable personality
traits.
- Using reprogenetics to target personality traits such as
agreeableness and so on is somewhat more concerning, to me, than disease
risks and intelligence.
- My guess is that it’s fine, but one reason for concern is that
there’s more potential for abuse by pushing to extremes. Adding 3.5 SDs
of IQ or -3.5 SDs of disease risk should be fine / safe. However, as a
simplified example to illustrate, adding 3.5 SDs of disagreeableness
might amount to creating something like a psychopath, and it’s not
extremely difficult to imagine some parents wanting to do that. Another
example would be “obedience”; parents, or even states, might want to
increase obedience pathologically.
- But I wonder if there are any traits that would have upside to
increasing, and no plausible appeal to decreasing. As an example, just
to illustrate, I would speculatively conjecture that there’s some kind
of trait like “gets the main first-order bits right; is pointed in the
right direction, along all the important dimensions” (see https://www.lesswrong.com/posts/fzKfzXWEBaENJXDGP/what-is-wisdom-1#Hypothesis).
- Are there any such traits?
- How can they be defined, measured, and polygenically predicted?
- Besides wisdom, what about empathy? Strategic ability? Sanity?
Reflectivity? Moral goodness? Ethicality? Responsibility? Integrity?
Honesty? Good judgement? Ability to coordinate?
- (lower priority, but interesting) Nonlinear and causal
models. What are the prospects for modeling nonlinear genome
-> phenome maps? What about modeling coarse causal structure of those
maps (such as “IQ affects EA but mostly not the other way around”)?
Might these models make better use of large heterogeneous databanks?
What about models with a mixture of circuit depths (linear, one
nonlinearity, two nonlinearities, etc.)? What about modeling all
available phenotypes at once? What about discovering latent
variables?
Given the massive size of the gene editing field, I doubt that there
are many “gap” projects I could recognize that would meaningfully
accelerate the relevant technologies; but I’m open to being
convinced.
Some projects I could be interested in assisting:
- If the task of editing a stem cell population many times in series
is very neglected, that could be relevant (though my guess is that
existing efforts such as Colossal would have this covered?).
- Methods to prevent excessive
DNA damage during culturing.
- Methods to make large insertions—both the technology and also
theoretical work analyzing the potential genomic vectoring power of that
operation. For example, what are the densest clusters of IQ-affecting
variants within one small region of a chromosome? With a realistic
stochastic model of an individual’s genome and resulting crossover
chromosomes, what is the expectation of post hoc densest region of
effect-weighted IQ-negative variants—i.e. what’s the best you can do by
replacing some 1-megabase region?
- Tests of strong genomic vectoring methods of any kind to greatly
increase polygenic traits in agricultural species (where the epigenomic
correctness problem can be largely ignored), in order to get end-to-end
feedback / validation on strong genomic vectoring.
