Identifying features of beneficial trials. (A). Relative influence of each selenium model parameter on trial outcomes as identified by the machine learning classifier. Higher relative influence indicates that trial outcomes are more sensitive to variation or uncertainty in the corresponding parameter. (B). Selected curves from Figure 3B for selenium model parameter sets predicted by the classifier to produce beneficial trial outcomes. Not all of these parameter sets produce beneficial outcomes because the classifier is not 100% accurate, but all harmful outcomes are eliminated and the proportion of beneficial outcomes is dramatically increased.