Figure 4
From: Real-time prediction of cell division timing in developing zebrafish embryo
Leave-one-out cross validation (LOOCV) of the sequential Bayesian inference method.
(a) Estimation of division timing. LOOCV was performed using the time-series dataset consisting of 39 individual cells. Results using sequential Bayesian inference (Bayes) and maximum likelihood estimator (MLE) methods are shown. Each line represents the results of a single cell. Diagonal line (green) represents the perfect prediction. (b) Compilation of the predictions shown at a. Average (red or blue dots for Bayes or MLE, respectively) and standard deviation (red or blue lines for Bayes or MLE, respectively) are shown for each time-point (time-frame number, τ). Diagonal line (green) represents the perfect prediction. (c) Comparison of sequential Bayesian inference and MLE methods. Prediction errors for sequential Bayesian inference (red bars) and MLE (blue bars) methods are shown as bar graph. Prediction errors are indicated by the number of time-point deviations from the perfect division-timing prediction (i.e., observed). The superior predictions with sequential Bayes are statistically significant as compared to those with MLE (p = 1.42e−21, Mann-Whitney U test).
