Fig. 5: A model-based approach with nested cross-validation reveals that the depth of a subset of lateral prefrontal tertiary sulci explains individual variability in reasoning above and beyond age.
From: Cognitive insights from tertiary sulci in prefrontal cortex
a Left: Spearman’s correlation between measured and predicted Matrix reasoning scores in the Replication sample (n = 27 participants) for the best tertiary sulci + age model, which includes the depths of the two most relevant sulci (pmfs-iRH + pimfsRH) from the Discovery sample, as well as age (Supplementary Fig. 5 for a model with all three tertiary sulci selected from the Discovery sample). Gray shading represents the 95% confidence interval for the linear model. Right: Density plot of model fit. The predicted scores from the chosen model (pmfs-iRH + pimfsRH + age) are shown in orange and overlaid on the distribution of measured Matrix reasoning scores (gray). b Distribution of predicted scores for the cross-validated nested model comparisons. Green: age only. Blue: all right hemisphere (RH) LPFC sulci + age. Each of the model fits is overlaid on the distribution of measured Matrix reasoning scores (gray). The pmfs-iRH + pimfsRH + age model (a) produced a better fit than both comparison models. Cross-validated mean-squared error (MSECV) and model fit (R2CV) are reported for each of the three models. c Empirical MSE for each of the three models estimated with a bootstrapping procedure (niterations = 10,000) to address the potential for leave-one-out cross-validation to result in high variance and overfitting. The model including all LPFC RH sulci + age (blue) exhibited notably high variance in error estimation. The red vertical line indicates the estimated median MSE. Source data are provided as a Source data file.
