Fig. 3. OMI variables accurately distinguish cells under low or high cardiomyocyte differentiation efficiency conditions on day 1.

a UMAP dimensionality reduction was performed on all 13 OMI variables (optical redox ratio, NAD(P)H τ_m, τ₁, τ₂, α₁, α₂, and intensity; FAD τ_m, τ₁, τ₂, α₁, α₂, and intensity) for each cell on day 1 and projected onto 2D space. Day 1 cells from all 15 conditions shown in Table 1 are plotted together with cells from low (< 50% cTnT+ on day 12) and high (≥ 50% cTnT+ on day 12) CM differentiation efficiencies in light gray and dark gray, respectively. n = 16048 and 14415 cells for low and high differentiation efficiency conditions, respectively. b–e All OMI data from day 1 cells were separated into two datasets. Dataset 1 was randomly partitioned into 80% portions for training and 20% portions for testing, respectively (n = 8974 cells for training, n = 2244 cells for test). Dataset 2 was used for evaluation of classifier performance. Binary classification was tested for low (< 50% cTnT+) vs. high (≥ 50% cTnT+) differentiation efficiency conditions on day 1. b OMI variable weights are shown specific to the logistic regression model. c Classification accuracy with respect to number of OMI variables was evaluated by chi-squared variable selection to separate low and high differentiation efficiency conditions with the logistic regression model. The number of variables included in the logistic regression model are indicated at bottom-axis. The accuracy scores are presented as mean ± STDEV. d The variables included for each logistic regression model [specified by numbers of variables on the x-axis in (c)] are defined, where the blue text indicates NAD(P)H lifetime variables and the red text indicates FAD lifetime variables. The OMI variables included in each instance (e.g., 3, 4) are indicated by a light blue + in each column. e Model performance of the logistic regression classifier was evaluated by receiver operating characteristic (ROC) curves using different OMI variable combinations as labeled. The area under the curve (AUC) is provided for each variable combination as indicated in the legend. Source data are provided as a source data file.