Characterization of the Sequence of Cytopenia after Total-body Irradiation using Unsupervised Machine Learning

Area of Science:

• Hematology
• Radiation Oncology
• Computational Biology

Background:

• Hematopoietic acute radiation syndrome (H-ARS) causes significant myelosuppression after total-body irradiation (TBI).
• Understanding the precise sequence of blood cell loss and recovery is crucial for developing effective countermeasures.
• Non-human primates (NHPs) serve as a relevant model for studying TBI-induced H-ARS.

Purpose of the Study:

• To characterize the temporal sequence of pancytopenia following TBI in male and female NHPs.
• To evaluate the impact of sargramostim treatment on blood cell recovery dynamics.
• To apply unsupervised machine learning for detailed analysis of longitudinal blood cell data.

Main Methods:

• Pooled longitudinal blood cell data from 501 NHPs across three randomized trials.
• Unsupervised machine learning, specifically hierarchical clustering with dynamic time warping (DTW), was used.
• Comparison of blood cell recovery trends between sargramostim-treated and vehicle-treated groups, stratified by sex.

Main Results:

• Lymphocytes and reticulocytes were the first cell types to deplete post-TBI, followed by WBC-related cells, platelets, and RBC-related cells.
• Nadir points occurred at distinct times: reticulocytes (day 8), WBCs (day 12), platelets (day 13), and RBCs (day 17).
• Sargramostim treatment led to earlier or similar recovery times compared to controls; females exhibited deeper nadirs than males.

Conclusions:

• Unsupervised ML provides a robust method for characterizing pancytopenia sequences in H-ARS.
• The findings elucidate the pathogenesis of H-ARS, informing therapeutic strategies and countermeasure development.
• Understanding sex-specific differences and treatment effects is vital for optimizing H-ARS management and medical triage.