I personally know the author of this paper who conducted these analyses. He is my office mate.

He discovered his own error that then led to this retraction. The error involved under-documented statistical code which expected the coding of male and female (0,1) to be reversed when training a classifier on each group, is my understanding. He kept the coding of sex the same across both training regimes. This created erroneous results. He became suspicious about the results after new applications of the analysis to new dataset produced over the top beautiful results.

The take is this. The mistake was honest, and way too easy. My office mate is a superb scientist with the highest integrity. He caught a mistake, and fessed up to it. This is good science.

Note: I am not part of the group that published this paper.