IDDF2023-ABS-0135 Construction of exosome transcriptomic signature for noninvasive and early detection of gastric cancer patients by machine learning: a multi-cohort study

Background

Gastric cancer (GC) is a common and preventable disease with a lack of accurate early diagnostic methods. Exosomal non-coding RNA (ncRNA), a class of liquid biopsy, has become a kind of promising diagnostic biomarker for many tumors. Our primary objective was to identify an ncRNA-based serum exosome signature for the GC diagnosis.

Methods

Serum exosomes from patients with GC (n=37) and healthy individuals (n=20) were characterized by RNA sequencing and potential biomarkers for GC were validated by qPCR (IDDF2023-ABS-0135 Figure 1. An overall work landscape of the whole study). From a modeling set of 522 GC patients and 460 healthy individuals, combined diagnostic score (cd-score) for GC were identified by machine learning-based algorithms and validated in an external validation set and predictive set. The receiver operating characteristic (ROC) curve was used to quantify the diagnostic capability of ncRNA-based cd-score by assessing areas under the curve (AUC), sensitivity and specificity.

Results

183 candidate biomarkers for GC were defined by RNA sequencing, and 31 of them were verified as potential biomarkers by qPCR. The cd-score, containing seven long ncRNAs (RP11-556O9.4, RP11-417E7.1, RPS3AP23, RP11-559M23.1, CTD-2339L15.3, DGCR9 and LINC00567) and one circular RNA (hsa_circ_0047880), was disclosed by applying multiple logistic regression for the diagnosis of GC and overpowers CEA, CA19-9 and CA72-4 alone or conjunctively. The cd-score could discriminate GC patients from healthy individuals with high accuracy in the training set (AUC=0.961, sensitivity=91.9%, specificity=90.0%), internal validation set (AUC=0.976, sensitivity=96.3%, specificity=87.0%), external validation set (AUC=0.939, sensitivity=100.0%, specificity=69.9%) and predictive set (AUC=0.987, sensitivity=93.1%, specificity=91.1%). Moreover, the cd-score allowed the diagnosis of GC versus precancerous lesions (PL, AUC=0.752, sensitivity=65.0%, specificity=72.6%). A series of bioinformatic and molecular biological assays unveiled that hsa_circ_0047880 was positively associated with the development of GC.

Conclusions

The ncRNA-based serum exosome cd-score for the prediction and early diagnosis of GC provides a promising liquid biopsy strategy for personalized medicine.