Table 3 Results of the clustering strategy on the predicted Q × E effects compared with the raw simulated Q × E effects and the clustering of the simulated Q × E effects
Number of QTL with simulated QEmbedded ImageE effectsPearson correlation rawaPPVb rawTPRc rawTop-10 rankdSpearman correlation rawPPV clusteredeTPR clusteredSpearman correlation clustered
50.725100%46.0%60%f0.67195.0%65.3%0.783
250.77799.4%42.0%78%0.63490.7%68.4%0.777
500.801100%47.6%74%0.68683.7%76.0%0.794
1000.81299.5%42.0%72%0.65584.5%72.4%0.780
2500.79697.4%41.3%54%0.64684.5%68.1%0.766
5000.81095.1%40.4%53%0.65980.2%66.1%0.745
  • The values in this table are averages over the 10 environments included in the trial because in each environment the QTL had different effects.

  • a Raw: without clustering on the simulated Q × E effects. The clustering of the predicted values is compared with the set of QTL with a truly simulated Q × E effect.

  • b PPV, positive predictive value, relative number of QTL in the cluster with the highest absolute predicted Q × E effects that are also in the cluster of QTL with a high simulated Q × E effect.

  • c TPR, true positive rate, relative number of QTL with a truly high simulated Q × E effect found in the cluster with the highest absolute predicted Q × E effects.

  • d Top-10 rank, relative number of the 10 QTL with the highest absolute simulated Q × E effect that are also predicted as being in the top 10 of most contributing QTL in that environment.

  • e Clustered, with clustering on the simulated Q × E effects. The clustering of the predicted values for the Q × E effects is compared with the cluster of QTL with the highest simulated Q × E effects.

  • f Top-5 rank instead of top-10 rank.