I have a dataset of 950 lipids (X) and want to see if any are correlated with cognitive function (Y). When I try to fit an opls regression model, it errors and says that "No model was built because the first predictive component was already insignificant".
However, when I loop through each of these lipids individually and perform a univariate correlation (Pearson/Spearman), I am left with ~30 lipids that are significantly correlated with cognitive function following FDR correction.
What does this mean? If there are univariate correlations, why am I not able to fit a multivariate model?