IDDF2024-ABS-0067 Construction and validation of a prognostic signature of pancreatic adenocarcinoma based on multiple machine learning algorithms associated with mitochondrial dynamics-related genes

<h3>Background</h3> Pancreatic adenocarcinoma (PAAD) is a highly aggressive and malignant cancer type characterized by few early symptoms and high mortality. Mitochondrial dynamics-related genes (MDRGs) can affect various processes during cancer development directly or indirectly. The area of MDRGs as biomarkers remains unexplored in PAAD. <h3>Methods</h3> The sequencing data and clinical information were from TCGA and GEO databases. The MDRGs were collected from MitoCarta, GO, and KEGG databases. The differentially expressed genes and overall survival-associated genes were selected with the R package ‘limma’ and univariate Cox regression. The prognostic signature was developed using lasso regression and multiple machine learning algorithms from the Python ‘scikit-survival’ library. The R package ‘CIBERSORT’ was used to evaluate the tumor microenvironment. The R package ‘oncoPredict’ was used to predict the IC50 score. The R package ‘maftools’ was used to visualize the somatic mutation. <h3>Results</h3> We obtained 37 genes through the intersection of differentially expressed genes (IDDF2024-ABS-0067 Figure 1. Identification of the 37 MDRGs (A)), overall survival-associated genes, and MDRGs (IDDF2024-ABS-0067 Figure 1. Identification of the 37 MDRGs (B)). GO and KEGG analysis of the 37 genes were performed (IDDF2024-ABS-0067 Figure 1. Identification of the 37 MDRGs (C-D)). A 12-gene prognostic signature was developed based on the 37 genes by multiple machine-learning algorithms (IDDF2024-ABS-0067 Figure 2. Construction of the machine learning signature (A)). The KM curves of the high-and low-risk groups demonstrated significant stratification in both the TCGA cohort and GSE62452 cohort (IDDF2024-ABS-0067 Figure 2. Construction of the machine learning signature (B-C)). Univariate and multiple Cox analysis validated the prognostic ability of the 12 genes and highlighted FAM73B as a potential biomarker (IDDF2024-ABS-0067 Figure 2. Construction of the machine learning signature (D-E)). The infiltration level of macrophages, monocytes, and CD8+ T cells showed significant differences between the high- and low-risk groups. The high-and low-risk groups exhibited distinct sensitivity to Gemcitabine and Paclitaxel (IDDF2024-ABS-0067 Figure 3. Immune infiltration, drug sensitivity, and somatic mutational features of high-and low-risk groups (A-D)). The somatic mutation between the groups was visualized in IDDF2024-ABS-0067 Figure 3. Immune infiltration, drug sensitivity, and somatic mutational features of high-and low-risk groups (E). <h3>Conclusions</h3> We developed an MDRG signature based on machine learning algorithms, providing new insights into the prognosis, drug sensitivity, tumor microenvironment, and somatic mutation for PAAD management.