Proposing new early detection indicators for pancreatic cancer: Combining machine learning and neural networks for serum miRNA-based diagnostic model

Abstract

BackgroundPancreatic cancer (PC) is a lethal malignancy that ranks seventh in terms of global cancer-related mortality. Despite advancements in treatment, the five-year survival rate remains low, emphasizing the urgent need for reliable early detection methods. MicroRNAs (miRNAs), a group of non-coding RNAs involved in critical gene regulatory mechanisms, have garnered significant attention as potential diagnostic and prognostic biomarkers for pancreatic cancer (PC). Their suitability stems from their accessibility and stability in blood, making them particularly appealing for clinical applications.MethodsIn this study, we analyzed serum miRNA expression profiles from three independent PC datasets obtained from the Gene Expression Omnibus (GEO) database. To identify serum miRNAs associated with PC incidence, we employed three machine learning algorithms: Support Vector Machine-Recursive Feature Elimination (SVM-RFE), Least Absolute Shrinkage and Selection Operator (LASSO), and Random Forest. We developed an artificial neural network model to assess the accuracy of the identified PC-related serum miRNAs (PCRSMs) and create a nomogram. These findings were further validated through qPCR experiments. Additionally, patient samples with PC were classified using the consensus clustering method.ResultsOur analysis revealed three PCRSMs, namely hsa-miR-4648, hsa-miR-125b-1-3p, and hsa-miR-3201, using the three machine learning algorithms. The artificial neural network model demonstrated high accuracy in distinguishing between normal and pancreatic cancer samples, with verification and training groups exhibiting AUC values of 0.935 and 0.926, respectively. We also utilized the consensus clustering method to classify PC samples into two optimal subtypes. Furthermore, our investigation into the expression of PCRSMs unveiled a significant negative correlation between the expression of hsa-miR-125b-1-3p and age.ConclusionOur study introduces a novel artificial neural network model for early diagnosis of pancreatic cancer, carrying significant clinical implications. Furthermore, our findings provide valuable insights into the pathogenesis of pancreatic cancer and offer potential avenues for drug screening, personalized treatment, and immunotherapy against this lethal disease.

Unraveling the role of disulfidptosis-related LncRNAs in colon cancer: a prognostic indicator for immunotherapy response, chemotherapy sensitivity, and insights into cell death mechanisms

“…Through comprehensive analysis of transcriptomic data, clinical data and mutation data of COAD tumor samples in the TCGA database, we successfully identified four DRLs and their significance in assessing the prognosis and immune status of COAD patients. Through one-way Cox and machine learning analyses, we identified AC083900.1, AP003555.1, SNHG7, and ZEB1-AS1 as independent prognostic variables for COAD ( Chi et al, 2023 ). By categorizing patients into low-risk and high-risk groups based on these DRLs, we achieved accurate categorization and demonstrated that their predictive performance was superior to traditional clinical indicators.…”

Revealing the diagnostic value and immune infiltration of senescence-related genes in endometriosis: a combined single-cell and machine learning analysis

“…The cohort of 45 DE-SRGs underwent a meticulous screening process whereing three distinct machine learning algorithms were harmoniously amalgamated: support vector machine-recursive feature elimination (SVM-RFE), random forest (RF), and the least absolute shrinkage and selection operator (LASSO) ( Chi et al, 2023b ; Song et al, 2023 ; Zhang et al, 2023 ). SVM-RFE, an advance upon the sequential backward selection algorithm rooted in the SVM’s tenet of maximal margin, delivers superior and more proficient classification performance, particularly for high-dimensional datasets.…”

Decoding tumor heterogeneity in uveal melanoma: basement membrane genes as novel biomarkers and therapeutic targets revealed by multi-omics approaches for cancer immunotherapy

“…To explore the potential prognostic relevance of these genes, univariate Cox regression analysis was performed, resulting in the identification of 81 genes significantly associated with survival outcomes. To further refine the gene set and mitigate the risk of overfitting, we employed the LASSO (Least Absolute Shrinkage and Selection Operator) method, a powerful machine learning approach ( Chi et al, 2023a ; Chi et al, 2023b ). The “glmnet” R package ( Engebretsen and Bohlin, 2019 ; Ren et al, 2023 ) was utilized to implement LASSO, which involves adding a penalty term to the regression model.…”