Greetings folks
We have performed a longitudinal study on human patients in response to disease treatment. The patient response is nonlinear (asymptotic) in time, and the data was not all measured at the same time points.
Our first attempt was to perform a piecewise linear GWAS (e.g. day 0-90, 90-180, 180-360, etc). The results were not good -- results from each time interval were highly correlated with the next, which indicates some sort of technical artifact (something like overfitting or not enough data points in each interval). Breaking it up into intervals like this also just isn't ideal from the standpoint that you're throwing out the data from outside each interval.
One package that looks nice for us is Chao Ning's GMA. https://academic.oup.com/bioinformatics/article-abstract/35/23/4879/5487383?redirectedFrom=fulltext They fit Legendre polynomials to the data, and test to see if any of the polynomial coefficient effect sizes differ from zero with SNP. It also incorporates a GRM, which is important to account for human genetic diversity. And they explicitly account for unbalanced data (not all of the data is taken at the same time points). But I've gotten stuck on basic things like trying to examine the baseline Legendre polynomial fits -- it would be nice to be able to see if the fits match the data. There have been a couple of other minor issues along the way. I've tried contacting them by email and haven't gotten any response, so I'm currently working my way through their source code. Their example data runs, so I'm pretty sure I can get it working in the end, but I'm wondering if there's something out there that would require less work on our end.
If anyone has worked on a nonlinear GWAS in the past and has some suggestions for pipelines or software packages that might be user-friendly, please let us know!
Thanks,
David
