Related Experiment Video
Updated: Jun 21, 2026
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Identifying predictive hematological biomarkers for radiation exposure by machine learning in mouse models.
Hang Chang1,2, Yiyan Yao3, Jared DeChant4
1Biological Systems and Engineering Division, Lawrence Berkeley National Laboratory, 1 Cyclotron Rd, Berkeley, CA, USA. hchang@lbl.gov.
Communications Medicine
|June 19, 2026
Summary
This study developed a Sparse Complete Blood Count (CBC) model to rapidly assess radiation exposure. The model shows promise for biodosimetry, though its accuracy varies with genetic background.
Area of Science:
- Radiation biology
- Computational biology
- Biomedical engineering
Background:
- Population-scale radiation exposure assessment is crucial for radiological emergencies but limited by slow, costly methods.
- Rapid and affordable screening tools are needed for biodosimetry, especially for low-dose exposures and delayed assessments.
- Inter-individual differences in radiation sensitivity complicate accurate exposure estimation.
Purpose of the Study:
- To develop and validate a predictive model for radiation exposure using complete blood count (CBC) data.
- To identify key CBC parameters for estimating radiation dose across different time points and genetic backgrounds.
- To assess the model's performance in diverse mouse populations, including genetically varied strains.
Main Methods:
- Analysis of CBC profiles from over 1151 mice exposed to varying X-ray doses (0.05-4 Gy).
- Development of a Sparse CBC model using sparse representation learning to identify informative CBC parameters.
- Validation through exhaustive cross-validation, a prospective cohort (431 animals), and a genetically diverse Collaborative Cross (CC) cohort (1720 animals).
Main Results:
- The Sparse CBC model demonstrated good performance (AUC, accuracy, sensitivity >80%) in retrospective and prospective cohorts.
- Model performance was modest in the genetically diverse CC cohort, with significant variation across different CC strains.
- Host genetic background was identified as a significant factor influencing the predictive accuracy of the radiation exposure model.
Conclusions:
- The Sparse CBC model effectively utilizes CBC data for radiation exposure estimation across various mouse cohorts, including genetically diverse populations.
- CBC-based biodosimetry offers a complementary tool for radiation exposure assessment.
- Genetic background significantly impacts the performance of CBC-based radiation exposure prediction models.
