Thinking of an appropriate way to impute 23&me data for a random individual.
1. Pre-compute models using MACH. I would do this for a variety of populations, e.g. HGDP, HapMap etc.
2. User comes to you with their data. Choose the closest matching population.
3. Use mini-mac (same website as above) to impute SNPs for the used using model from step1.
I need to do some tests to see how accurate a model from a different population will be, but reckon if the population is a close enough match it should work.
It would be nice to set this up as a service to 23&me customers...let me know if interested!