LASSO and PCA
To determine the minimal number of antibody features that distinguish
between different groups, a least absolute shrinkage and selection
operator (LASSO) feature reduction method was employed as previously
described37. Cross validation was performed
iteratively (repeated 1,000 times; 10-fold cross validation) to identify
the optimal value of the regularized parameters. Unsupervised principal
component analysis (PCA) was then performed on LASSO-selected antibody
features (which resolves multiple variables into principal components
that describe the variance within the data set). The contribution of
each variable in describing the variance within each principal component
is represented on loading plots.