[eweitz]

Large genetic datasets yield medical progress by increasing statistical power of tests. These better tests enable earlier and more targeted treatment.

Take heart disease. It has a significant but complex genetic component. Many genetic variants each contribute a small amount to risk for heart disease. If a given person has many small risk variants, the sum total risk -- often called "polygenic score" -- can be relatively high.

People in the top 8% of polygenic scores had a 3x higher risk for heart disease than the general population [1][2]. Through techniques like polygenic scoring, large genetic datasets enable uniquely early detection of high risk for the world's leading cause of death.

[1] https://www.nature.com/articles/s41588-018-0183-z

[2] https://www.vox.com/science-and-health/2018/8/24/17759772/ge...

[maxheadroom]

Ah, yes, polygenic scoring and the heart disease increase rate. Did you happen to catch this[0] refutation of the single source's work?

[0] - https://twitter.com/cecilejanssens/status/103135930540723404...

Also, you're posing prediction and targeted treatment but you haven't posited how Bob's mapped genome sitting in 23andme will be used for medical treatment.

As we know, genes are not an emphatic, "this will happen to you," but an increase in likelihood; which still doesn't translate to any emphatic treatments from the genes, themselves, yeah?

[eweitz]

> refutation of the single source's work

Janssens seems less skeptical of 23andMe's paper on polygenic score for type 2 diabetes [1][2], which -- interestingly -- positively cites the Khera 2018 paper on polygenic score for heart disease that she critiqued. Some researchers are skeptical, but the medical community generally seems to consider polygenic scores promising for tests [3][4].

> you haven't posited how Bob's mapped genome sitting in 23andme will be used for medical treatment.

Early intervention. Polygenic scores could be used for medical treatment by motivating earlier intervention. That could include stronger recommendations for better diet and exercise, closer monitoring programs, or more precise prescriptions. That, in turn, could reduce disease burden.

[1] https://twitter.com/cecilejanssens/status/113707970323438797...

[2] https://permalinks.23andme.com/pdf/23_19-Type2Diabetes_March...

[3] https://twitter.com/EricTopol/status/1129780543434964993

[4] https://journals.plos.org/plosmedicine/article?id=10.1371/jo...

[maxheadroom]

>...could be used for medical treatment by motivating earlier intervention...

and

>...could include stronger recommendations for better diet and exercise, closer monitoring programs, or more precise prescriptions...

and

>...could reduce disease burden...

This is where the problem delineates for me: We're being massively assumptive in moving from "could" to "is" and "will".

I will, generally, concede the could portion but to assert that it is emphatically happening or going to happen is still far from fruition and to label this science as such, just yet, is overreaching and giving false hope where none should really be given because, then, you'll taint it's benefits with the drawbacks.

Remember: Anonymised data (e.g.: 23andme) only allows a survey of what's relatively known or can be inferred from the anonymised dataset.

To arrive at what you're suggesting, it would have to move into a different realm (I believe), like UK BioBank or GEDMatch but, even then, we're still basing things on speculative science - gambles of percentages that aren't, emphatically, true or false but a kind of "maybe, kind of, sort of, in a way, definitely could or defintely could not" muddied waters.

That, to me, is a far stretch from saying that the data in 23andme is - actually - helping medicine; which I believe is what the OC I replied to emphatically said.