Genomic prediction for soybean segregating populations: selection strategies and training set establishment