ML-based mobilization score predicts response to G-CSF-mediated stem cell mobilization from pre-mobilization data. (A) Mobilization score derived from the Random Forest algorithm-based ML model vs Day 1 PB CD34 count in G-CSF mobilized donors (N = 798). (B) Distribution of “poor mobilizer” (Day 1 PB CD34 cells <40/μL), “less-than-optimal mobilizer” (Day 1 PB CD34 cells 20-40/μL) and “good-mobilizer” (Day 1 PB CD34 cells >40/μL) based on their mobilization score calculated by Random Forest based ML model (N = 798). (C) Confusion matrix, PPV, NPV and accuracy of the prediction in all patients (N = 799) from the Random Forest-based model. One hundred twenty-seven out of 171 “poor/less-than-optimal mobilizers” and 536 out of 628 “good mobilizers” were correctly predicted. (D) Mobilization score derived from the AdaBoost algorithm vs Day 1 PB CD34 count in G-CSF mobilized donors (N = 798). (E) Distribution of “poor mobilizer” (Day 1 PB CD34 count <40/μL), “less-than-optimal mobilizer” (Day 1 PB CD34 count 20-40/μL) and “good mobilizer” (Day 1 PB CD34 count >40/μL) based on their mobilization score calculated by Adaboost based ML model. (F) Confusion matrix, PPV, NPV and accuracy of the prediction in all patients (N = 799) from the AdaBoost-based model. 128 out of 171 “poor/less-than-optimal mobilizers” and 531 out of 628 “good mobilizers” were correctly predicted.