Skip to main content

Table 3 Performance (correlation coefficients and rmse) of our models in comparison to ShiftX2 using the test set created by our pipeline

From: NightShift: NMR shift inference by general hybrid model training - a framework for NMR chemical shift prediction

Prediction N correlation CA correlation CB correlation C correlation H correlation
Method (rmse) (rmse) (rmse) (rmse) (rmse)
Spinster 0.817 (2.977) 0.956 (1.425) 0.992 (1.582) 0.731 (1.524) 0.593 (0.505)
ShiftX2 0.554 (5.606) 0.953 (1.475) 0.984 (2.238) 0.711 (1.65) 0.534 (0.583)
Training / test size 39,147 / 26,099 39,947 / 26,632 36,211 / 24,142 29,065 / 19,377 41,076 / 27,385
Prediction HA correlation HB correlation HD correlation HEHZ correlation HG correlation
method (rmse) (rmse) (rmse) (rmse) (rmse)
Spinster 0.997 (2.889) 0.82 (0.559) 0.994 (0.324) 0.981 (0.375) 0.86 (0.365)
ShiftX2 0.517 (5.998) 0.721 (1.033) 0.012 (0.706) 0.816 (0.505) -0.147 (0.967)
Training / test size 42,639 / 28,427 37,263 / 24,843 22,508 / 15,006 12,919 / 8,613 25,932 /17,289
  1. The size is measured in the number of available atomic shifts.