# A Pancreatic Ductal Adenocarcinoma Diagnostic System Using Serum Extracellular Vesicle Detection with Optimized Lectin Combination Using Machine Learning

## TL;DR

A new blood test using machine learning and specific sugar markers on cell fragments could help detect pancreatic cancer early.

## Contribution

A novel diagnostic system for PDAC using serum EV glycan markers and machine learning with optimized lectin combination.

## Key findings

- The lectin combination Jacalin and ABA achieved high diagnostic performance (AUC = 0.890 and 0.971) for PDAC detection.
- The system showed an AUC of 0.870 for stage I PDAC patients, indicating potential for early detection.
- The system outperformed the conventional marker CA19.9 (AUC = 0.752).

## Abstract

This study aimed to establish a novel diagnostic system for pancreatic ductal adenocarcinoma (PDAC) by identifying extracellular vesicles (EVs) with specific glycan markers in the blood using a highly sensitive EV-counting system that we previously developed. We performed a multiplex assay using lectins that recognize specific glycans on EVs in the serum. The glycan alteration signature of serum EVs from PDAC patients was analyzed using machine learning (support vector machine), resulting in the identification of an optimal lectin combination, Jacalin and Agaricus bisporus agglutinin (ABA), that achieved high diagnostic performance of PDAC. This lectin-based system, reflecting changes in Jacalin/ABA binding, demonstrated significantly higher diagnostic performance (area under the curve [AUC] = 0.890 and 0.971). Notably, the system achieved an AUC of 0.870 in patients with the stage I disease. These findings highlight the potential of a serum EV-based diagnostic system leveraging Jacalin and ABA glycan recognition for the early detection of PDAC.

Background: Pancreatic ductal adenocarcinoma (PDAC) has one of the poorest prognoses among malignant tumors, mainly due to the difficulty of early diagnosis. Therefore, it is crucial to identify reliable blood markers for a highly sensitive diagnostic system. We previously developed a highly sensitive extracellular vesicle (EV)-counting system, which can quantify the absolute number of specific EVs in serum. In this study, a multiplex assay using lectins that recognize specific glycans on EVs in the serum of PDAC patients was performed to select the optimal lectin combination. Methods: The glycan alteration signature of serum EVs from patients with PDAC was analyzed using a lectin-based multiplex assay combined with the EV-counting system. The optimal lectin combination that recognizes PDAC-specific changes was selected using machine learning analyses (support vector machine) for high diagnostic performance across independent patient cohorts. Results: An optimal lectin combination, Jacalin and Agaricus bisporus agglutinin (ABA), for PDAC detection was identified using machine learning analysis. This lectin-based system, reflecting changes in Jacalin/ABA binding, showed significantly higher diagnostic performance (area under the curve [AUC] = 0.890 and 0.971) than that of the conventional diagnostic marker carbohydrate antigen 19-9 (CA19.9; AUC = 0.752). Notably, the system achieved an AUC of 0.870 in patients with the stage I disease. Conclusions: These findings highlight the potential of a serum EV-based diagnostic system leveraging Jacalin and ABA glycan recognition for the early detection of PDAC.