A Robust Diagnostic Signature For The Pancreatic Cancer Based On Within-sample Relative Expression Orderings

Pancreatic ductal adenocarcinoma is one of the most aggressive malignant neoplasms with poor outcomes.But the diagnosis of pancreatic cancer is very difficult.Current diagnostic methods have some limitations.Imaging examinations such as Computed Tomography(CT),Magnetic Resonance Imaging(MRI),Endoscopic Ultrasonography(EUS)are greatly influenced by the examining equipment and experience of doctors.Some commonly used molecular markers,such as CA19-9 and KRAS gene mutation have a insufficient sensitivity or specificity.Fine needle aspiration biopsy combined with pathological examination is the gold standard for the diagnosis of pancreatic cancer,but its diagnostic sensitivity is low due to inaccurate aspiration location and insufficient aspiration volume.Furthermore,the operation is difficult and it's easy to cause serious complications because the concealed location and abundant blood vessels of pancreas.In addition,the development of DNA chip and high-throughput sequencing technology has produced a large number of pancreatic tissue expression profile data,many studies use these data to identify quantitative transcriptional markers of pancreatic cancer based on gene expression levels.However,such markers are difficult to convert into clinical use.Many studies have shown that qualitative transcriptional markers based on relative expression orderings(REO)are more stable,which are not sensitive to batch effects and may be applicable to multiple platforms.The purpose of our study was identified a transcriptome qualitative diagnosis marker of pancreatic cancer,which can be used in cases when the sampling location is inaccurate or the sampling amount is insufficient,and can making a effectivelydistinguish between non-cancer and pancreatic cancer and suitable for gene expression profile data of biological samples from different platforms.We used five sets of data from the Affymetrix platform and the Illumina platform for training,including 49 normal samples,60 chronic pancreatitis samples,326 surgically resected pancreatic ductal adenocarcinoma samples,and 145 adjacent samples of cancer.Normal and chronic pancreatitis are combined into a non-cancerous group and compare to pancreatic ductal adenocarcinoma sample,calculating 85% highly stable reversal pairs with a total of 2557 pairs.The majority voting rule is used to calculate F_score,then we get a diagnosis marker with the highest F_score,including 6 gene pairs.In 8 sets of independent datasets in the public database(all datasets contains surgical resection,two of them contains biopsy samples),for the surgically resected samples,the accuracies for normal,chronic pancreatitis,pancreatic ductal adenocarcinoma and cancer-adjacent samples were: 95.83%(23/24)),100%(11/11),96.34%(316/328)and 98.33%(59/60),respectively.For biopsy samples,the accuracy in TCGA's 28 PDAC samples was 100%(28/28)and GSE42952's 23 PDAC biopsy samples was 95.65%(22/23).The markers we identify can accurately identify pancreatic and non-cancer samples in multiple platforms from three different testing companies: Affymetrix,Agilent,and Illumina.Additionally,our markers identify the adjacent tissues as cancer accurately,it also can be used in biopsy data.Therefore,our markers may assist current biopsy diagnosis of pancreatic cancer to improve the diagnostic accuracy,and may solve the wrong diagnosis of pancreatic cancer.