Bridging the Generalization Gap: Training Robust Models on Confounded Biological Data