Fig. 3

Cell fate prediction based on genes selected by correlation analysis. a Highly correlated genes with cell death. These genes were extracted by using Spearman’s rank correlation analysis approach from 100 training datasets (10 times of 10-fold cross-validation), with the top 30 genes highly correlated with cell death selected in each training dataset. b, c Prediction accuracies of cell death by using different degree polynomial models (linear, quadratic and cubic polynomials) in (b), as well as quadratic polynomial model with different number of correlated gene pairs in (c). d, e Stability comparison of different models. For each model, a total of 10,000 randomly selected points were used to measure its stability. After 10 times of 10-fold cross-validation, 100 regression values were obtained for each point. Then the variance of each point can be derived. We use the MVAV (mean value of the 10,000 variances) to assess the instability of each model. Thus, smaller value of MVAV indicates that the model is more stable. The bold markers denote that, within the corresponding model, the 10,000 variances obey gamma distribution