4 ms·
This line of thinking really highlights how building models that predict phenotype from genotype assuming linear and independent genetic contributions, though u
by mxwsn 8y ago
This line of thinking really highlights how building models that predict phenotype from genotype assuming linear and independent genetic contributions, though useful for predicting something like height [0] [1], is not going to generalize sufficiently to solve the lofty goal of personalized precision medicine even with an explosion in data.
Biologically, we know tons and tons about the interconnected biochemical pathways that cause a lot of activity in the cell. It's impressive that statistical modeling approaches that ignore this knowledge earned by thousands of humanyears worth of work can work so well in some domains, but I'm wary of the possibility of academic communities isolating themselves from empiricism and continuing to make mathematically-convenient assumptions, which in the long-run helps no one but themselves.
[0] https://news.ycombinator.com/item?id=16392877 https://news.ycombinator.com/item?id=16392877
[1] https://www.biorxiv.org/content/early/2017/09/19/190124 https://www.biorxiv.org/content/early/2017/09/19/190124
- carbocation 8y agoI don't think this is ignored by the scientific community at all. For example, Eric Lander's group has a pretty nice treatment of why additive models can't explain heritability [1]. And this is coming from the same Eric Lander who has done some of the key work in complex disease / additive genetic modeling. Imagine you're in a world where you are handed down some knowledge there are Newtonian models, and more advanced models that use relativity, but you don't know what the constants are yet. We're still looking for the Newtonian constants. We know there is much more out there once we understand the basics. But the basics are still worth understanding. 1 = http://www.pnas.org/content/109/4/1193 http://www.pnas.org/content/109/4/1193