Fig. 2. The RPN-signature predicts individual pain sensitivity based on pain-free resting-state functional brain connectivity.

The learning-curve (a) suggests that the size of the training sample (Study 1) was sufficient to substantially reduce overfitting and improve generalisation. Internal cross-validated prediction in the training sample (Study 1, N = 35, b) and prospective external validation in the test samples (Studies 2 and 3, c, d, N = 37 and 19, respectively) revealed considerable predictive accuracy, robustness, and multicentre generalisability of the RPN-signature. Mean absolute error (MAE) of the prediction is depicted by dashed lines. Shaded ribbons imply the 95% confidence intervals for the regression estimates, Pearson-correlation (r) of the predicted vs. observed values and the corresponding permutation-based p-value is given. Source data are provided as a Source Data file.