On the predictive power of sequence similarity in yeast