We present a computationally efficient workflow with the following steps:
1. Comparing a broad panel of predictive models via cross validation and selecting good ones for multiple traits
2. Using the selected models to evaluate the full potential predictive distribution of the traits across all crosses
3. Assessing the predictive distributions to select good crosses both visually and algorithmically and finding optimal compromises across traits
4. Simulating virtual progeny from the selected crosses
5. Cycling back to Step 2 and monitoring improvement
We use real data from an oat trial for illustration and JMP Genomics software for each step.