NEWS

AffiXcan 1.7.1

Null eigenvectors for TBA with null variance or NA values in the training set are returned correctly

For each fold, for each gene for which a GReX could be imputed, the following values are now returned
Squared correlation of the model's predictions with training data
P-value of the model
Corrected p-value of the model with benjamini-hochberg procedure
Correlation of the model's predictions with validation data
Squared correlation of the model's predictions with validation data
P-value of the correlation test of the model's predictions with validation data

P-value of the cor.test() between predicted GReX and real expression values of genes is returned in when performing cross-validation
When using affiXcanTrain in cross-validation mode, three main values for each gene are therefore returned: rho and rho squared (see changes in v 1.3.1), and the p-value of cor.test()

Providing population structure covariates is now not mandatory to perform models training

ANOVA p-value < 0.05 to assess prediction significance is not used anymore; instead:
A k-fold cross-validation on the training dataset may be performed; k can be defined by the user
Pearson correlation coefficients (R) and determination coefficients (R^2) between predicted GReX and real expression values of genes are computed
In literature, GReX of genes for which the mean of the R^2 is above 0.01 are considered non-randomly predicted, according to the new benchmarking standards [ref]