Multidimensional proteomics and explainable AI feature selection identify cross-platform lung cancer molecular signature in blood plasma.

BACKGROUND: Lung cancer is the leading cause of cancer mortality worldwide despite the availability of low-dose computed tomography (LDCT) for screening in high-risk populations. METHODS: To develop an approach and identify blood-based protein signatures for lung cancer that can be deployed across platforms, we combined data-independent acquisition mass-spectrometry (DIA-MS) and proximity extension assay (PEA) with explainable artificial intelligence (XAI)-led machine learning (ML) for plasma-based biomarker discovery. Using a cohort of 490 lung cancer patients and 124 matched controls, ML models were trained to predict lung cancer and XAI was used to characterise networks of model-consistent features. We then introduced a DNA-aptamer based proteomic approach to assess cross-platform concordance and define a cross-platform signature. This signature was subsequently evaluated using an external cohort. RESULTS: Here we show that ML models achieve an AUROC of 0.91 [95% CI: 0.88-0.93] and 0.97 [95% CI: 0.92-0.98] in DIA-MS and PEA, respectively, using a 80/20% train/holdout split. XAI further characterises networks of model-consistent features related to chemotaxis, cell adhesion, wound healing and immune response. Introduction of the DNA-aptamer proteomic approach identifies a cross-platform signature, with performances of 0.88 [95% CI: 0.80-0.90] and 0.88 [95% CI: 0.81-0.95] in DIA-MS and PEA, respectively. Assessment of this signature in an external cohort separates lung cancer from control cases. CONCLUSIONS: This study develops an approach combining multi-dimensional proteomics with XAI-ML and demonstrates the characterisation of cross-platform biomarker signatures for lung cancer.