For the risk prediction, when we encounter an ambiguous SNP e.g. ref = A, alt = T, we will just flip it so that it will be the same as the one in the p-value file. Might need to figure out a better way to do that in the future

For risk prediction, we haven't implement a way to deal with missing data. Currently we directly set the genotype*effect as 0. However, we do know that this will still affect the decomposition as we are trying to solve the whole system of equation
This will be something that we need to do in the future

Currently we don't take in any information from the user as to what the reference Strand should be. Will need this for better calculation of LD (might need to flip the reference and target)

For LD block size, we currently only allow the maximum and minimum block size set for ALL chromosomes. Might want to allow people to do it individually in the future.
Also, we are using the densest region as the estimated block size. Dynamic block size might be something good?

Similar to the variance problem, when performing case control analysis, we don't actually allow a different sample size for each individual SNPs e.g. all SNPs will assume having the same case and same control.

Our current variance estimation only use the maximum sample size. So if there are 2 SNPs in the data, one with 1000 samples, and the other have 10 samples, the variance calculated will be using 1000 samples

When trying to perform risk prediction, we need to compute the standardized genotype. However, if there is no variation in the sample (e.g. all sample has the same genotype), we cannot standardize them as the SD = 0.

Currently we just set the genotype to 0. A better way might be to pre-filter those genotype just like what we've done with the LD matrix. Which one is better?