Fig. 2: Spatial ageing clocks.

a, Pipeline for building spatial ageing clocks from the coronal section dataset and using them to compare ageing across conditions or quantify the deviation (age acceleration) of a cell from its expected predicted age. b, Predicted age as a function of actual age, with predicted ages obtained by leaving out all cells from one mouse and training spatial ageing clocks on the remaining data to make predictions on the held-out cells and repeating this procedure for all mice. Heat map shows density of predicted ages, circles show median predicted age per mouse, and the line of best fit for the median predicted ages is shown in black. Pearson correlation between predicted age and actual age for all cells is reported as R (with values of R > 0.7 in bold), and the Pearson correlation between median predicted age and actual age for all mice is reported as r; 95% confidence intervals for correlations are shown in brackets. c, Dot plot comparing the performance of single-cell ageing clocks without spatial smoothing (SingleCell) with our spatial ageing clocks (SpatialSmooth). Dot colour corresponds to Pearson correlation and dot size is inversely related to mean absolute error between predicted age and actual age. d, Density of predicted ages in an external 140-gene MERFISH dataset consisting of 6 coronal sections from 3 mice (3, 19 and 25 months old). e, Density of predicted ages across the young (<9 months) and old (>19 months) sagittal section samples. f, Summary of the training data used to build the spatial ageing clocks and the generalization of spatial ageing clocks to different datasets. scRNA-seq, single-cell RNA-seq; snRNA-seq, single-nucleus RNA-seq.