Root mean squared error (RMSE) measures prediction error as the square root of the mean squared differences between predicted and observed values. Because the square root reverses the squaring step, RMSE is expressed in the target’s original units: an MSE of 2.29 becomes an RMSE of 1.51 log-molarity units. section 4.1.4
Use RMSE to interpret regression performance in practical units:
- Compute predictions on a held-out validation set, then take the square root of mean squared error. section 4.3.1
- Compare the result with a baseline rather than treating the score alone as meaningful; an ESOL validation benchmark has RMSE 1.4579 under the original coefficients. section 4.3.1
- Remember that larger errors receive greater weight because the differences are squared before averaging. section 4.1.4
Drug-target affinity
For DGraphDTA on the DAVIS dataset, an MSE of 0.202 gives RMSE √0.202 ≈ 0.45 pKd units. Since pKd is logarithmic, that corresponds to typical errors of roughly 2.8-fold in dissociation constant and helps distinguish high-affinity from weak binders. section 11.6.2
Inspect the mean squared error equation before applying its square root to obtain RMSE. section 4.1.4









