Open Source Genetic Screening
Can you open source pre-implantation genetic screening for embryos? Has this already occurred? Also, not a medical guy, totally a data guy, would appreciate correction for errors in either.
So embryo selection is something I’ve kept my eye on ever since Gwern’s classic essay on the subject (1). For those of you who haven’t read about it, the core concept is that, during artificial insemination, instead of creating, like, 3-4 viable embryos, you create 30-50 and then run modern genetic testing to predict their life outcomes. Gwern was really interested in intelligence amplification but I’m interested in something more mundane and…practical, which is predicting and preventing diseases. This is already done, pretty automatically, for diseases where a single gene is responsible; that gene is busted, therefore the baby will have X disease, therefore don’t implant that embryo. What’s new, and caught a bunch of attention in Israel recently, is diseases caused by a complicated mixture of genes (2). So you take the embryo’s genome, run it against a machine learning algorithm trained to predict the disease, and then select the genome with the lowest risk of that disease.
So my question, simply, is whether there’s any medical or technical issue preventing me or anyone else from using a service like Lifeview (3), which does a bunch of this kind of screening but not for particular diseases, asking them for a copy of the genomes, then running an algorithm I found off Github to predict for a specific disease I know I’m high risk for, and then requesting certain embryos from Lifeview based on that?
Let’s get specific and look at psoriasis, since this is a rabbit hole I’ve fallen down. Psoriasis is a complicated disease that basically makes your skin scaly and, in about a third of cases, develops into psioratic arthritis. Now there’s existing algorithms out there in the medical literature (4) and (5). One of these is 99.8% accurate in predicting psoriasis. And there is, literally, a psoriasis prediction algorithm on Github (6), you can go download the python files now. Now I haven’t worked with genetic data before but it’s fundamentally just data and, if nothing else, there’s an R library for it somewhere. So let’s say someone had psoriasis, they didn’t want their child to inherit it, and they either had some data science chops or they hired someone like me. Could/would Livewell send us, say, 40-50 pandas dataframes with the embryo’s genetic data, and then they/I could run this algorithm off github, identify which embryos were at highest and lowest risk of psoriasis, and then send that back to Livewell?
At the technical level, this seems…like a weekend project. Like, you’d need to get used to the data and I’m sure there’s some specific libraries you’d need to get used to but, fundamentally, grab this standardized dataset and apply this pre-existing algorithm to it is kinda grunt work. And for people who think it sounds crazy reckless to genetically engineer your child based on random code you found on the internet, I have bad news for you regarding virtually every piece of software you interact with on a daily basis, including medical software. Github is the good stuff, most of the time we’re slapping together stuff we found on Stack Overflow. I guarantee your local hospital runs software built like this. But it’s the medical level I’m unsure of. It…feels like there’s a lot of room for unexpected side effects.
So, at a gut level, what is preventing this from being done at this moment, from either a technical, medical, or even legal level?
And this seems wild but I have a condition like this and I’m sure there’s plenty of people here in a similar situation. For example, Freddie DeBoer has written movingly on his struggles with bipolar disorder and he probably has the money and skills to execute something like this, and certainly would probably wish to spare his prospective children from inheriting his condition. I’m sure there’s plenty of people here with worries about autism or other…um…a typical neural conditions or physical disabilities they wouldn’t want to pass on to their children. And this is also a highly technically literate group with plenty of techy money.
But, before running down all the implications, I’m really just trying to get a feel for whether this is possible and what the limitations are. Because it seems like something “potentially” really big and doable but…I mean, sometimes you find $20 laying on the ground but I’m always a bit suspicious.
(1) https://www.gwern.net/Embryo-selection
(3) https://www.lifeview.com/our_tests.html
(4) https://www.sciencedirect.com/science/article/pii/S0888754313002048
(5) https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3022824/
(6) https://github.com/genomicdatascience/psopredict