Background. There is a poor prognosis for diffuse large B-cell lymphoma (DLBCL), one of the most common types of non-Hodgkin lymphoma (NHL). Through gene expression profiles, this study intends to reveal potential subtypes among patients with DLBCL by evaluating their prognostic impact on immune cells. Methods. Immune subtypes were developed based on CD8+ T cells and natural killer cells calculated from gene expression profiles. The comparison of prognoses and enriched pathways was made between immune subtypes. Following this validation step, samples from the independent data set were analyzed to determine the correlation between immune subtype and prognosis and immune checkpoint blockade (ICB) response. To provide a model to predict the DLBCL immune subtypes, machine learning methods were used. The virtual screening and molecular docking were adopted to identify small molecules to target the immune subtype biomarkers. Results. A training data set containing 432 DLBCL samples from five data sets and a testing dataset containing 420 DLBCL samples from GSE10846 were used to develop and validate immune subtypes. There were two novel immune subtypes identified in this study: an inflamed subtype (IS) and a noninflamed subtype (NIS). When compared with NIS, IS was associated with higher levels of immune cells and a better prognosis for immunotherapy. Based on the random forest algorithm, a robust machine learning model has been established by 12 hub genes, and the area under the curve (AUC) value is 0.948. Three small molecules were selected to target NIS biomarkers, including VGF, RAD54L, and FKBP8. Conclusion. This study assessed immune cells as prognostic factors in DLBCL, constructed an immune subtype that could be used to identify patients who would benefit from ICB, and constructed a model to predict the immune subtype.

Background. The epithelial mesenchymal transition (EMT) gene has been shown to be significantly associated with the prognosis of solid tumors; however, there is a lack of models for the EMT gene to predict the prognosis of AML patients. Methods. First, we downloaded clinical data and raw transcriptome sequencing data from the TCGA database of acute myeloid leukemia (AML) patients. All currently confirmed EMT-related genes were obtained from the dbEMT 2.0 database, and 30% of the TCGA data were randomly selected as the test set. Univariate Cox regression analysis, random forest, and lasso regression were used to optimize the number of genes for model construction, and multivariate Cox regression was used for model construction. Area under the ROC curve was used to assess the efficacy of the model application, and the internal validation set was used to assess the stability of the model. Results. A total of 173 AML samples were downloaded, and a total of 1184 EMT-related genes were downloaded. The results of univariate batch Cox regression analysis suggested that 212 genes were associated with patient prognosis, random forest and lasso regression yielded 18 and 8 prognosis-related EMT genes, respectively, and the results of multifactorial COX regression model suggested that 5 genes, CBR1, HS3ST3B1, LIMA1, MIR573, and PTP4A3, were considered as independent risk factors affecting patient prognosis. The model ROC results suggested that the area under the curve was 0.868 and the internal validation results showed that the area under the curve was 0.815. Conclusion. During this study, we constructed a signature model of five EMT-related genes to predict overall survival in patients with AML; it will provide a useful tool for clinical decision making.

Background. Oxidative stress (OS) can either lead to leukemogenesis or induce tumor cell death by inflammation and immune response accompanying the process of OS through chemotherapy. However, previous studies mainly focus on the level of OS state and the salient factors leading to tumorigenesis and progression of acute myeloid leukemia (AML), and nothing has been done to distinguish the OS-related genes with different functions. Method. First, we downloaded single-cell RNA sequencing (scRNAseq) and bulk RNA sequencing (RNAseq) data from public databases and evaluated the oxidative stress functions between leukemia cells and normal cells by the ssGSEA algorithm. Then, we used machine learning methods to screen out OS gene set A related to the occurrence and prognosis of AML and OS gene set B related to treatment in leukemia stem cells (LSCs) like population (HSC-like). Furthermore, we screened out the hub genes in the above two gene sets and used them to identify molecular subclasses and construct a model for predicting therapy response. Results. Leukemia cells have different OS functions compared to normal cells and significant OS functional changes before and after chemotherapy. Two different clusters in gene set A were identified, which showed different biological properties and clinical relevance. The sensitive model for predicting therapy response based on gene set B demonstrated predictive accuracy by ROC and internal validation. Conclusion. We combined scRNAseq and bulk RNAseq data to construct two different transcriptomic profiles to reveal the different roles of OS-related genes involved in AML oncogenesis and chemotherapy resistance, which might provide important insights into the mechanism of OS-related genes in the pathogenesis and drug resistance of AML.