Embryo Size/Stage. The PLSR calibration and validation performance for phase 1 models based on the nanostring data (n = 96 embryos) is shown in Figure 5 (LV = 2). A PLSR model for embryo size calibrated to two-thirds of the data [R2 = 0.63, RMSE = 0.131 log10(cell number), n = 62] has a validation R2 of 0.58 validated on the remaining one-third of data [RMSE = 0.135 log10(cell number), n = 31, slope = 0.84 ± 0.13, intercept = 0.31 ± 0.25 log10(cell number), bias = −0.03 log10(cell number)]; that is, the majority of the variance in embryo size [log10(cell number)] is predictable based on the gene expression profile (3 embryos had missing size information). The small reduction in prediction performance for embryo size between calibration (R2 = 0.63) and validation (R2 = 0.58) is expected when we switch from training to testing data and the validation prediction performance remains acceptable. The validated model slope and intercept are not significantly different from 1 and 0, respectively, and the bias is close to 0. This indicates that the PLSR model can provide valid predictions of embryo cell number.The confusion matrix for the PLSR predictions of embryo stage (M, TM, EB, B, XB) are shown in Table 2. The model correctly classifies embryo stage in 19 out of 31 embryos for the validation data set. This is similar to the LDA model (Table 3), which correctly classifies embryo stage in 18 out of 31 embryos for the validation data set (although a direct comparison cannot be directly made because the LDA model is calibrated to the categorical embryo stage variable, whereas PLSR is calibrated to the continuous embryo size variable (Bereton and Loyd, 2014). The confusion matrix for the PLSR predictions of embryo grade and stage (e.g., B1, B2, and B3) for validation (n = 31) data sets are listed in Supplemental Table S2 (https:// doi .org/ 10 .3168/ jds .2017 -14306). The PLSR models correctly classify embryo grade and stage in 6 out of 31 embryos for the validation data set. The RMSE for the PLSR embryo size model [RMSE = 0.135 log10(cell number)] is approximately double the difference between grades and stage [~0.07 log10(cell number)] and the model provides poor prediction of grade and stage. The predictive performance of LDA models of embryo grade and stage were similar to the PLSR models with 8 out of 31 correct classifications of embryo grade and stage.