Explore the headline sword refers to the "king of cancer"!

"Pancreatic ductal adenocarcinoma (PDAC)" is one of the most lethal types of cancer, which is characterized by rapid progression, metastasis fast, difficult early diagnosis, recurrence rate, known as the "cancer of the King"
.

Many patients are already in the advanced stage of the disease when they are diagnosed with PDAC, miss the opportunity for surgical treatment, and the five-year survival rate is extremely low
.

There are three main reasons why PADC is difficult to diagnose early.
First, the anatomical location of the pancreas makes the tumor hidden and difficult to be found
.

Second, the patient’s early symptoms (such as weight loss, fatigue, abdominal and back pain and discomfort) are not specific enough to make a clear diagnosis
.

Furthermore, the existing non-invasive detection methods for pancreatic cancer are not yet mature
.

Therefore, it is necessary to find a more effective method to detect PDAC
.

Metabolomics can collect, detect and analyze various small molecular metabolites that are highly sensitive to biological activity and pathological conditions.
Accurate, robust and low-cost metabolomics detection methods provide hope for future disease diagnosis
.

In recent years, more and more studies have used artificial intelligence methods to analyze omics data and establish appropriate and effective detection or verification models for disease diagnosis, classification and efficacy prediction
.

The combined diagnosis of machine learning (ML) and metabolomics is currently a very attractive and promising concept, but previous work has mainly focused on model construction, rather than selecting key metabolites for disease detection
.

Recently, the team of Professor Yin Yuxin of Peking University School of Basic Medicine and collaborators of the Chinese Academy of Sciences and the Chinese People’s Liberation Army General Hospital applied machine learning combined with lipidomics and multi-omics techniques to comprehensively analyze the metabolic characteristics of pancreatic ductal adenocarcinoma (pancreatic cancer), and developed an artificial The intelligent-assisted PDAC serum metabolism detection method achieved 86.
74% and 85.
00% classification detection accuracy in a large external verification cohort of more than 1,000 cases and a prospective clinical cohort containing benign pancreatic lesions, and its detection efficiency was significantly better than CA19- 9 Check with CT
.

The research titled "Metabolic detection and systems analyses of pancreatic ductal adenocarcinoma through machine learning, lipomics, and multi-omics" was published online in Science Advances on December 22, 2021
.

Research results (Source: Science Advances) In most medical applications, ML methods are usually evaluated on a data set
.

In contrast, the ML-assisted metabolic PDAC detection method has been tested and evaluated by a large external validation cohort (n = 1003), demonstrating the stability of the method's performance
.

The characteristics of fast processing speed and high precision make this PDAC detection method have good application potential in the future
.

Traditionally, data dimensionality reduction and biomarker screening of metabolomics or lipidomics are mainly based on analysis of variance (ANOVA) and least square discriminant analysis (PLS-DA)
.

This research innovatively applied the greedy algorithm based on support vector machine (SVM), which showed excellent performance in the feature selection of serum lipidomics data
.

Tests on 1033 PDAC patients at different stages found that the accuracy of this method was 86.
74% in the large external validation cohort, the area under the curve (AUC) was 0.
9351, and the accuracy in the prospective clinical cohort was 85.
00%, and the AUC was 0.
9389
.

The ROC curve of the ML-assisted metabolite PDAC detection method in the validation study training data set & internal validation data set & external validation data set & prospective clinical cohort (Source: Science Advances) There are 17 types in the selected characteristic metabolites Lipids, including 4 types of lysophosphatidylcholine (LPC), 7 types of phosphatidylcholine (PC), 3 types of sphingomyelin (SMs), 1 type of lysophosphatidylethanolamine (LPE), 1 type of phosphatidylethanolamine ( PE) and 1 type of diglyceride (DG)
.

LPC, PC and PE are involved in glycerophospholipid metabolism, and SMs are involved in sphingolipid metabolism
.

Tissue proteomics and single-cell sequencing analysis showed that the metabolic pathways of glycerophospholipids and sphingolipids were disrupted in PDAC cells
.

A series of changes in these metabolites may reflect changes in lipid metabolism and related signal transduction pathways, proliferation and apoptosis resistance of cancer cells during the initiation and development of PDAC
.

In this study, the ML analysis of serum lipidomics, tissue proteomics, single-cell sequencing and other technologies are combined to characterize the lipid metabolism characteristics of PDAC from the perspective of the integration of peripheral circulation blood and tissue spatial lipidomics
.

Ion chromatograms of selected 17 characteristic lipid metabolites (Source: Science Advances) This work established a method combining metabolomics with ML and greedy algorithms, and used ML to refine the targeted metabolomics disease detection program
.

Except for CA19-9, there is no liquid-based detection method available for PDAC diagnosis
.

However, the detection of CA19-9 also has obvious limitations.
For example, patients with benign pancreaticobiliary disease may also have elevated CA19-9 due to biliary obstruction, which can easily be misdiagnosed as pancreatic cancer
.

The ML-assisted metabolic PDAC detection method is accurate, highly sensitive, minimally invasive (serum-based) and non-radioactive, which may help clinicians to make more comprehensive and accurate PDAC diagnosis and follow-up treatment
.

Therefore, incorporating it into current diagnostic methods may complement the routine diagnostic procedures for high-risk patients with PDAC
.

"Of course, there are some limitations of this study
.

" The features selected by the model cannot distinguish the early or late stages of PDAC, nor can it be used to predict the prognosis of PDAC patients
.

And this method is mainly based on the East Asian population, whether it is applicable to the detection of PDAC in other populations remains to be further studied
.

The use of metabolomics data must also consider the relationship between other metabolic diseases such as obesity, diabetes and PDAC, otherwise the performance of the ML-assisted metabolic PDAC detection method may be affected by metabolic-related confounding factors
.

This method should also be combined with existing detection methods such as CA19-9, abdominal ultrasound, CT, etc.
, to cautiously interpret the PDAC screening and detection results
.

End reference materials: [1] Wang G, Yao H, Gong Y, et al.
Metabolic detection and systems analyses of pancreatic ductal adenocarcinoma through machine learning, lipidomics, and multi-omics.
Sci Adv.
2021 Dec 24;7(52) :eabh2724.
doi: 10.
1126/sciadv.
abh2724.
Epub 2021 Dec 22.
PMID: 34936449.