Abstract
Abstract
Accurate differentiation among asthma, bronchiectasis, and chronic obstructive pulmonary disease (COPD) remains a major clinical challenge due to overlapping symptoms and limitations of conventional diagnostic approaches. This study proposes a nonparametric cumulative distribution function (CDF)-based framework combined with permutation testing to analyze volatile organic compounds (VOCs) in exhaled breath as non-invasive biomarkers for respiratory disease characterization. Breathomics data obtained via gas chromatography–mass spectrometry (GC-MS) from 121 patients were analyzed following log₂ transformation of VOC intensities. The analytical scope included not only shared VOCs but also pair-specific and disease-exclusive compounds. The discriminative ability of each VOC was assessed by comparing class-specific empirical cumulative distributions with null distributions generated through 2000 random permutations. CDF rankings were then generated for core, pair-specific, and disease-exclusive singleton VOCs. To evaluate classification utility, progressively reduced VOC subsets were assessed using Random Forest and Partial Least Squares Discriminant Analysis (PLS-DA) under repeated stratified cross-validation. Results showed that a reduced set of highly ranked VOCs preserved substantial classification performance, supporting the effectiveness of the proposed prioritization strategy. Candidate VOCs were then suggested based not only on their ranking scores, but also on biological interpretation and support from previously reported literature. Overall, the CDF- and permutation-based approach provides an interpretable, statistically robust, and model-independent framework for biomarker discovery in respiratory diseases.