Utilizing Deep Machine Learning to Disclose Hematological Indices and Biomarkers in Asthma-COPD Overlap

<bold>Background:</bold> Asthma-COPD overlap (ACO) combines both asthma and COPD features, and clues suggest that they undergo worse events than those either one alone. <bold>Methods:</bold> Using a dataset of hematological parameters collected retrospectively by deep machine learning to analyze from infants to nineties aged patients with ACO (n = 907), asthma (n = 31929), COPD (n = 15934), and healthy cases (n = 5181) in Southern China. This study focused on the potential differences in disease severity among asthma-COPD overlap, asthma, and COPD, as well as their effects on blood biochemical indicators, lung function indications, airway inflammation indicators, and lung microbial community. Bayesian and random forest prediction models were used to achieve visual monitoring of these diseases. <bold>Results:</bold> Through training and analysis with the deep learning model, we successfully identified biomarkers closely associated with the disease conditions of asthmatic and COPD patients from a vast amount of hematological data, such as FeNO and eosinophil count. These markers are involved in various biological processes, including inflammatory responses, immune regulation, and oxidative stress. Importantly, we discovered that these markers exhibit significant changes in the early stages of the diseases, suggesting their potential role in early diagnosis. <bold>Conclusion:</bold> Our study demonstrates for the first time that through the use of deep learning, we can mine hematological biomarkers related to asthma and COPD, such as FeNO and eosinophil count, from large-scale blood data. These markers offer new insights for clinical diagnosis and treatment.