Figure 4.

A Chronological age classifier for human pancreatic cells. (a) A single chronological age classifier for the entire ensemble of human pancreatic cells using machine learning. No cell-type segregation was performed during training. (b) Barplot showing the accuracy of GERAS on classifying the age of pancreatic cells that were not used for training the model. An accuracy of 95% was achieved for cells previously unseen by GERAS. (b’) The classification accuracy of GERAS on the previously unseen pancreatic cells after segregating them into major cell-types. Classification accuracy equals the proportion of cells for which the classification stage matched the actual stage. For each cell-type, greater than 93% accuracy was achieved. Error bars indicate standard error. F₁-scores, a measure of precision and sensitivity of the classifier, are depicted at the bottom. (c) External validation for the classifier was provided by human pancreatic single-cell mRNA expression data obtained from an independent publication. Cells from individuals belonging to the ‘Middle’ (38–54 years) stage of the classifier displayed greater than 93% accuracy. (d) Balloonplot showing classification of cells from individuals with similar chronological age but different BMI. In individuals with normal BMI, 32% of the cells were classified in ‘Juvenile’ and ‘Young’ stages, while none (0%) of the cells from individuals with obese BMI were similarly classified. Number of cells for each condition is denoted by ‘n’.