Fig. 5
From: Machine learning identifies a compact gene set for monitoring the circadian clock in human blood
Personalized prediction of CT in control samples from the three datasets. All predictors were trained using sumabsv = 2 and nSPC = 2. Levels of statistical significance (paired Wilcoxon rank-sum test): * P < 10–3, ** P = 7.4 × 10–5, *** P = 5.9 × 10–11. a Schematic of procedure for training personal and ensemble predictors with universal guidance. For the training sets, the height represents observations and the width represents features (e.g. genes). Universal guidance refers to filtering for only those genes used by the universal predictor. b Boxplots of absolute error for universal (standard tenfold cross-validation), personal (leave-one-sample-out cross-validation for each individual), and ensemble (circular mean of universal and personal) predictors. c Boxplots of median absolute error (by individual) for universal, personal, and ensemble predictors