An autoantibody-based machine learning classifier for the detection of early-stage non-small cell lung cancer

The humoral immune system plays a significant role in the immune response to cancer but is challenging to study at scale. We used programmable phage immunoprecipitation sequencing (PhIP-Seq) to profile the autoantibody repertoire in non-small cell lung cancer (NSCLC) patients for the purpose of training a machine learning-based classifier to distinguish NSCLC patients from healthy controls using 301 primarily early-stage, asymptomatic NSCLC patients and 352 healthy controls. The classifier performed well in cross-validation (average ROC-AUC = 0.94) and in an independently analyzed clinical validation cohort of 134 NSCLC patients and 96 healthy controls (ROC-AUC = 0.84). Classification performance can be maintained with only a few hundred target peptides, provided a sufficiently large cohort is used for optimal training. Our findings suggest the existence of a measurable autoreactive humoral profile in NSCLC and demonstrate the potential for serum-based early detection of cancer independent of nucleic acids.