Fig. 2: Machine learning modelling results for pathogen attachment using Random Forest.

a Plot of the predicted versus measured average attachment values for one run of the P. aeruginosa test set, b Regression model performance metric results for P. aeruginosa training and test sets, c P. aeruginosa descriptors ranked by their importance from top to bottom and how their value (high = yellow or low = purple) impacts on the model (positive or negative impact), d Feature coverage descriptor distribution against attachment for P. aeruginosa, e Scatter plot of the measured against predicted average attachment values for one run of the S. aureus test set, f Regression model performance metric results for the S. aureus training and test sets, g S. aureus descriptor importance and their impact on model output. The topographical descriptors found to be most important for bacterial attachment are the inscribed circle which relates to the space between primitives, the total area covered by features (feature coverage) and the maximum feature radius, h Feature coverage descriptor distribution against attachment for S. aureus, i Comparison between P. aeruginosa attachment and the engineered roughness index (ERI) calculated for the topographies investigated, j Comparison between S. aureus attachment and the roughness index calculated for the topographies. The source data are provided as a Source Data file.