Maybe, just maybe, we will be able to find some of the epistatic genetic effects through clever search algos. https://arxiv.org/abs/1610.05108 pic.twitter.com/KlPgz0fIOh
-
Or we can just chuck deep learning at it. :) I've suggested dilated CNNs for multi-trait GWAS and any available annotations. Handles arbitrarily complex interactions...
At the cost of huge variance. I think we are still in the p/n region where it makes sense to strongly direct the model wrt. bias variance trade-off.
-
-
I dunno about that. p>>n is precisely where NNs tend to shine empirically, and they should benefit from modeling the distribution of effect sizes, genetic correlations, higher-order interactions, and everything a linear model or something like xyz throws away. Empirical question.
-
Depends on how complex genomic causation is. If it's mostly linear, then using NNs should not result in improvements over modeling with strong assumptions (glm models), but as you say, empirical question. Eventually, NNs will be superior as n -> inf.
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.