Integrating multi-omics and machine learning survival frameworks to build a prognostic model based on immune function and cell death patterns in a lung adenocarcinoma cohort

Abstract

Introduction: The programmed cell death (PCD) plays a key role in the development and progression of lung adenocarcinoma. In addition, immune-related genes also play a crucial role in cancer progression and patient prognosis. However, further studies are needed to investigate the prognostic significance of the interaction between immune-related genes and cell death in LUAD.

Methods: In this study, 10 clustering algorithms were applied to perform molecular typing based on cell death-related genes, immune-related genes, methylation data and somatic mutation data. And a powerful computational framework was used to investigate the relationship between immune genes and cell death patterns in LUAD patients. A total of 10 commonly used machine learning algorithms were collected and subsequently combined into 101 unique combinations, and we constructed an immune-associated programmed cell death model (PIGRS) using the machine learning model that exhibited the best performance. Finally, based on a series of in vitro experiments used to explore the role of PSME3 in LUAD.

Results: We used 10 clustering algorithms and multi-omics data to categorize TCGA-LUAD patients into three subtypes. patients with the CS3 subtype had the best prognosis, whereas patients with the CS1 and CS2 subtypes had a poorer prognosis. PIGRS, a combination of 15 high-impact genes, showed strong prognostic performance for LUAD patients. PIGRS has a very strong prognostic efficacy compared to our collection. In conclusion, we found that PSME3 has been little studied in lung adenocarcinoma and may be a novel prognostic factor in lung adenocarcinoma.

Discussion: Three LUAD subtypes with different molecular features and clinical significance were successfully identified by bioinformatic analysis, and PIGRS was constructed using a powerful machine learning framework. and investigated PSME3, which may affect apoptosis in lung adenocarcinoma cells through the PI3K/AKT/Bcl-2 signaling pathway.

Keywords: immunotherapy efficacy; lung adenocarcinoma; machine learning; precision medicine; programmed cell death.

Figure 1

Flowchart of this study.

Figure 2

Identification of three subtypes and assessment of clinical prognosis of different subtypes. (A) Veen plot demonstrating 20 cell death genes; (B) multi-omics characterization of the three subtypes; (C) OS survival prognosis of the three subtypes; (D) PFI prognosis of the three subtypes.

Figure 3

Assessment of mutation status and pathway analysis across different subtypes. (A) Veen plot demonstrating 20 cell death genes; (B) multi-omics characterization of the three subtypes; (C) OS survival prognosis of the three subtypes; (D) PFI prognosis of the three subtypes.

Figure 4

Assessment of the immune microenvironment in different subtypes. (A) Heatmap showing the expression of immune checkpoint genes and the level of tumor-infiltrating lymphocytes in each subtype; (B) Expression levels of CTLA4 and PDCD1; (C-E) IC50 values of commonly used chemotherapeutic agents; (F) Heatmap of NTPs generated according to subtype-specific up-regulation of biomarkers identified in the LUAD cohort; (G) km survival curves of the META cohort.

Figure 5

Multiple machine learning constructs for PIGRS. (A) Heatmap showing 101 machine learning; (B) Lasso screening genes; (C) GBM algorithm showing the importance of different genes; (D-G) km survival curves for high and low PIGRS (D) TCGA; (E) GSE31210; (F) GSE50081; (G) META cohort.

Figure 6

Model comparisons. (A) Evaluation of PIGRS for predicting 1-, 2-, 3-, 4-, and 5-year survivability in patients; (B-E) PIGRS compared with published models.

Figure 7

Functional enrichment analysis of the PIGRS group. (A) GSVA enrichment analysis in the PIGRS group; (B) KEGG enrichment analysis in the PIGRS group; (C, E) GSEA enrichment analysis in the high PIGRS group; (D, F) GSEA enrichment pathway in the low PIGRS group.

Figure 8

Genomic variation and intra-tumor heterogeneity in different PIGRS subgroups. (A) Violin plots demonstrating TMB differences between the high and low PIGRS groups; (B) correlation between TMB and PIGRS; (C) Kaplan-Meier curves analyzing OS by combining the TMB score and PIGRS risk score. (D) Violin plot showing the difference in mutant allele tumor heterogeneity (MATH) scores between the high PIGRS and low PIGRS groups. (E) Correlation between MATH and PIGRS; (F) Kaplan-Meier curves analyzing OS by combining MATH score and PIGRS risk score. (G) Mutation analysis of the high PIGRS group. (H) Mutation analysis of the low PIGRS group. ***p < 0.001.

Figure 9

Exploration of the tumor immune microenvironment. (A) Stroma score, immunity score, ESTIMATE score, and tumor purity were used to quantify the different immune statuses between the high and low risk groups; (B) Heatmap demonstrating the situation of the six immune infiltration algorithms; (C) Differential expression of the various immune checkpoints in the high- and low-risk groups; (D) Differences in TIDE expression. *p<0.05,**p<0.01,***p < 0.01,****p < 0.0001.

Figure 10

Predicting and validating the immunologic efficacy of PIGRS. (A) IPS scores of high and low PIGRS groups; (B) Survival curves of high and low PIGRS groups in the IMvigor210 cohort. (C) Box line plot depicting the difference in risk scores between CR/PR patients and SD/PD patients in the IMvigor210 cohort. (D, E) km curves for the high and low PIGRS groups in the IMvigor210 staging. (F) Proportion of CR/PR or SD/PD patients receiving immunotherapy in the high and low risk groups of the IMvigor210 cohort. (G) Survival curves for high and low PIGRS in the GSE78220 cohort. (H) Proportion of patients with CR/PR or SD/PD who received immunotherapy in the high and low PIGRS groups in the GSE78220 cohort. (I) Box line plot depicting the difference in risk scores between CR/PR patients and SD/PD patients in the GSE78220 cohort. (J) The TIDE algorithm predicts response to immunotherapy between the high and low ERGRS groups. (K) Proportion of R or NR patients receiving immunotherapy in the high and low PIGRS groups of the TCGA-LUAD cohort. (L) Submap algorithm predicting response to immunotherapy between the high and low PIGRS groups. ns, p >0.05,*p < 0.05,***p<0.001,****p < 0.0001.

Figure 11

Validation of PIGRS markers. (A) t-SNE plot showing the cell types identified by marker genes. (B) Heatmap showing the 5 most important marker genes in each cell cluster. (C, D) Bubble map (C) and violin map (D) showing the enrichment scores of the PIGRS gene set for each cell type using AUCell, UCell, singscore, ssGSEA, and AddModulescore scores for the enrichment of the PIGRS gene set for each cell type. (E-G) Enrichment analysis of GSEA in the high PIGRS.Score group including Ferroptosis pathway (E), Necrosis pathway (F), Apoptosis pathway (G).

Figure 12

PSME3 affects the biological behavior of LUAD cells in vitro. (A) RT-qPCR to detect the expression of PSME3 mRNA. (B) The expression of PSME3 protein was detected by Western Blot. (C) Detection of β-actin, PSME3, AKT, p-AKT (Ser473), cleaved PARP, and Bcl-2 in PSME3 knockdown-treated A549 and H1299 cells by Western Blot. (D, F) RT-qPCR to detect the efficiency of Si-PSME3 transfection. (E, G) Growth curves of PSME3 knockdown-treated A549 and H1299 cells were determined using CCK8. (H, I) Cell scratch assay to detect the invasive ability of A549 and H1299 cells after PSME3 knockdown treatment. (J, K) Transwell assay to detect the invasion ability of A549 and H1299 cells after PSME3 knockdown treatment. (L, M) Colony formation assay was performed to detect the proliferation of A549 and H1299 cells. (“***” indicates p < 0.001).