Fig. 5: The morphome reliably predicts nanotopography-induced gene expression.

a The morphome was used to train a Bayesian linear regression model that predicted myogenic, osteogenic, chondrogenic, and fibrotic gene expression. The expression of each gene (response) was trained against linear combinations of morphome features (predictors) and without any prior knowledge on topography parameters. The linear regression model was trained using 60% of the dataset and tested using 40% of the data. Scatterplots show actual and predicted gene expression values by using the test set as input to the model, with each open faced circle showing predicted and actual gene expression from a single morphome. Black diamond shows the median predicted gene expression values. Colors represent the different musculoskeletal genes, with orange denoting myogenic, blue denoting osteoblastic, green denoting chondrogenic and brown denoting fibrotic genes. Mean absolute error (MAE) was obtained by first calculating the difference between actual and predicted gene expression values for each morphome then averaging differences across the entire morphome. b Testing the predictive power of the morphome by leave-one-out validation. To test the predictive power and bias of the morphome, the linear regression model was retrained after exclusion of one combination of cell type and topography. The excluded cell type and topography dataset was used for prediction, from which MAE was calculated. The tile position denotes the cell type and nanotopography combination that was excluded in the model and used for testing, while the color of each tile denotes the MAE.