Fig. 5: BATLI analyses combined with machine learning allows the prediction of replication outcomes in single-cells before the on-set of replication.

A Predictive accuracy across different trajectory lengths. The bar plot shows the percentage of correctly predicted true fates (y-axis) for both replicative (red) and non-replicative (blue) macrophages across different trajectory lengths (x-axis, hpi) used for training (x-axis, representing hours post-infection, e.g., “5” means data from 0 to 5 hpi were used). B Confusion matrix representing the performance of the logistic regression classification model in predicting whether human macrophages infected with L. pneumophila support bacterial replication or not, using the 0–5 hpi range of times as training datasets. Matrix compares the true cell fates (Y-axis) with the predicted fates (X-axis), measured as percentages. C Logistic regression equation describing the probability (P) of a single infected macrophage supporting L. pneumophila replication. This probability is derived from Δψm (TMRM SD/mean) and mROS (CellROX SD/mean) delta-encoded values at 5 hpi. D Receiver Operating Characteristic (ROC) curve for the logistic regression model predicting replicative vacuoles in L. pneumophila-infected macrophages. The X-axis represents the false positive rate (1-specificity), while the Y-axis represents the true positive rate (sensitivity). AUC Area Under Curve. E Precision-recall curve showing model performance, with the logistic regression model (blue curve) compared to a no-skill model (gray dashed line) and a hypothetical perfect model (red dashed line).