jellily / pcaseq Goto Github PK

Allow the user to control how many loci are read in at the same time (currently set to 5000(

Make the weights parameter take a function and provide the Beta weight function as one built-in option. This means that the beta weight function needs to be added in to the name space, and the functions that take weights need to be modified. Probably also want to include a function to find the MAF as a convenience for users of the package.

The weighting function should take in a MAF and return a numeric value that is not Inf or NaN.

Add testing for these two functions. Currently the functions they call are tested.

When the MAF is NA, an error triggers--something is wrong with the logic.

Currently, seqPCA returns the SNPs in the data set, before filtering. Need to return the actual SNPs used to calculate the GRM. (requires changing the GRM calculation function to return the SNPs actually used).

Conditions to check:

1. Letters instead of numbers
2. Data isn't in 0,1,2 format
3. Non-matrix (or at least cannot be converted to matrix)
4. Missing genotypes
5. Only one subject
6. Only one locus

Looks like missing values are ignored unless they are filtered out---double check this

Make these test cases more comprehensive.

Add an argument to PopStructure that takes a vector of length 2 that sets the allele frequency of loci to keep in the data set.

Look into how this is done in SNPRelate and SeqArray/SeqVarTools

Allow both SeqArray data structures and use the functions from SeqArray/SeqVarTools

Make the code that finds the GRM automatically remove monomorphic and singleton loci to improve computation time.

Allow the user to select a lower bound on the MAF; primarily for use with EIGENSTRAT method.

Currently, for each block of 5,000 SNPs, the SNP filtering functions throw an error if there is no remaining SNPs after filter. However, it's possible in a data set to get one block that has no remaining SNPs but still have a valid data set. Instead, for each block, there should be a warning to the user and after all the filtering is finished, check once to make sure there is still SNP data (and if not throw an error).

currently returns NULL

Verify that if a SNP has missing data, that the final GRM entry is not missing and is properly calculated. Create a test case for this.