Early detection of pancreatic cancer by comprehensive serum miRNA sequencing with automated machine learning

Abstract

Background: Pancreatic cancer is often diagnosed at advanced stages, and early-stage diagnosis of pancreatic cancer is difficult because of nonspecific symptoms and lack of available biomarkers.

Methods: We performed comprehensive serum miRNA sequencing of 212 pancreatic cancer patient samples from 14 hospitals and 213 non-cancerous healthy control samples. We randomly classified the pancreatic cancer and control samples into two cohorts: a training cohort (N = 185) and a validation cohort (N = 240). We created ensemble models that combined automated machine learning with 100 highly expressed miRNAs and their combination with CA19-9 and validated the performance of the models in the independent validation cohort.

Results: The diagnostic model with the combination of the 100 highly expressed miRNAs and CA19-9 could discriminate pancreatic cancer from non-cancer healthy control with high accuracy (area under the curve (AUC), 0.99; sensitivity, 90%; specificity, 98%). We validated high diagnostic accuracy in an independent asymptomatic early-stage (stage 0-I) pancreatic cancer cohort (AUC:0.97; sensitivity, 67%; specificity, 98%).

Conclusions: We demonstrate that the 100 highly expressed miRNAs and their combination with CA19-9 could be biomarkers for the specific and early detection of pancreatic cancer.

Fig. 1. The workflow of pancreatic cancer patients and healthy control participants for developing and evaluating prediction models.

a All the 425 serum samples, including 213 pancreatic cancer patients from 14 hospitals and 212 healthy controls without cancer from three clinics, were randomly divided into the training and validation cohorts.

Fig. 2. The serum miRNA expression profiles in healthy controls differed from those in pancreatic cancer patients.

a Heatmap representing hierarchical unsupervised clustering analysis of the miRNA sequencing of healthy controls (blue) and pancreatic cancer patients (orange). b Principal component analysis (PCA) of the miRNA sequencing of healthy controls (blue) and pancreatic cancer patients (orange). c–e The PCA grouped by hospitals, clinics, and the NGS measurements. HC healthy control, PC pancreatic cancer.

Fig. 3. Creation of the best discrimination models using AutoML in the training cohort and validation in the independent validation cohort.

a Schematic procedure of the index calculation in miRNA model and miRNA+CA19-9 model. b ROC curve for the performance of serum CA19-9 alone (red), miRNA model (green), and miRNA+CA19-9 model (blue) in the validation cohort. c–e Box plots of CA19-9, indices of miRNA model, and miRNA+CA19-9 model of healthy participants and pancreatic cancer patients in each stage (0, I, II, III, and IV) in the validation cohort. f, g Confusion matrices of miRNA model and miRNA+CA19-9 model at 98% specificity in each stage of the validation cohort. TP True Positive, TN True Negative, FP False Positive, FN False Negative, SEN Sensitivity, SPC Specificity, FPR False Positive Rate, FNR False Negative Rate. The color gradient indicated the rate of each metric. h, i Box plots of CEA and DUPAN-2 of healthy participants and pancreatic cancer patients in each stage (0, I, II, III, and IV) in the validation cohort.

Fig. 4. The 100 highly expressed miRNAs and CA19-9 in blood serum successfully discriminated healthy controls from patients with pancreatic cancer, not only in the advanced stage but also in the early stage (stage 0-I) with high sensitivity and specificity.

a, b ROC curves for the performance of serum CA19-9 alone (red), miRNA model (green), and miRNA+CA19-9 model (blue) in discriminating patients with pancreatic cancer in stage 0-I (a) and stage 0-II (b) from healthy controls. c ROC curves for the performance of serum CA19-9 alone (red), miRNA model (green), and miRNA+CA19-9 model (blue) in discriminating asymptomatic patients with pancreatic cancer in stage 0-I in the validation cohort. d–l Asymptomatic patients with pancreatic cancer in stage I: case 1 (d–g) and case 2 (h–l). d MRI revealed main pancreatic duct dilatation (arrow). Scale bar, 3.0 cm (e) MRI revealed a low-signal area in the main pancreatic duct on T2-weighted images (arrow). Scale bar, 3.0 cm (f) Contrast-enhanced CT revealed a slightly enhanced tumor in the main pancreatic duct (arrow). Scale bar, 3.0 cm (g) EUS revealed a mass, 5 mm in diameter, in the main pancreatic duct (arrow). Scale bar, 0.50 cm (h) MRI revealed main pancreatic duct dilatation (arrow). Scale bar, 3.0 cm (i) Contrast-enhanced CT revealed main pancreatic dilatation (arrow). A tumor could not be detected by CT. Scale bar, 3.0 cm (j) A tumor could not be detected in PET-CT (arrow). Scale bar, 3.0 cm (k) A tumor, 6 mm in diameter, was detected by EUS (arrow). Scale bar, 1.0 cm (l) Hematoxylin, and eosin staining of the surgical specimen revealed pancreatic cancer in stage IA. The inset shows invasive pancreatic cancer. Scale bar, 200 µm