10 plus tumor genome grouping on the new.

Today, I would like to share with you an article published on clinical cancer research in April 2020.this paper innovatively combines deep learning model, pan cancer genomics and tumor immunity, classifies tumors according to the differences of different tumor genome levels, and reveals its unique immune gene status and response to immunotherapy.from the current trend, deep learning and machine learning are more and more widely used in the field of bioinformatics.this article uses deep learning to analyze genomic sequencing data, as well as analysis of immune cell infiltration and tumor microenvironment.multi factor deep learning reveals that Pan cancer genomic tumor clusters have unique immune gene status and response to immunotherapy Abstract in this study, the authors developed a pan cancer deep machine learning model integrating tumor mutation load, microsatellite DNA instability and somatic cell copy number change. The 8646 tumors of 29 cancer types in TCGA were divided into 4 types And explore the characteristics of immune microenvironment of each genome group, and the relationship between each genome group and immunotherapy response.the analysis of RNA SEQ data revealed the unique immune microenvironment associated with each genome group.the model was applied to exome sequencing data from two immunotherapy clinical cohorts, which proved that tumor patients with different genome clusters had significantly different responses in immunotherapy. Br / >in order to verify the effect of different immune models in patients with melanoma, the results of immunotherapy from two groups of patients with melanoma were verified.cancer types in different genome groups (GC) have different effects on immunotherapy.results interpretation 1. The deep learning model integrates three major tumor genomic variation features (which are reported to be associated with immune checkpoint blocking reactions in a variety of cancer types). 1. MSI burden: the microsatellite instability burden value (expressed as the total number of MSI changes) 2 Mtmb: 1 29 tumor types 8646 from TCGA database The total numerical deep learning model of non synonymous mutations of samples consists of a mode specific deep belief network (DBN) and a self encoder (DAEs) for partitioning analysis. First, the deep belief network (DBN) is used to extract the high-level representation of each genome feature, and then the deep automatic encoder (DAEs) is performed to stratify the combined representation. Secondly, the tumors were divided into four groups (GC) by constructing a deep learning model.figure 1b shows significant differences in MSI burden, scna burden and mtmb among the four genomes.mtmb is closely related to traditional TMB (the total number of non synonymous mutations), while the correlation among mtmb, MSI burden and scna burden is weak to moderate, which indicates that each of these three genomic features may have a significant impact on tumor biology.our model stratified all tumor samples into four GC's (Fig. 1b and table S2).GC1 and GC3 appear to be genetically stable, characterized by low or medium mtmb, MSI and scna burden (later named GC1: tlowmlowslow or GC3: tlowmlowsmid).on the contrary, GC2 and GC4 are generally genetically unstable, with high mtmb, high MSI burden and low scna burden in GC2 (later named GC2: timhislow), while in GC4 there are high MSI burden, high scna burden and low mtmb (later named GC4: tlowmhishi).among the four GC categories, tumors in GC2 had the highest MSI burden (median = 722), the highest mtmb (median = 105), and GC4 tumors had the highest scna burden (median = 0.26).the tumors in GC1 had the lowest mtmb (median = 4) and MSI burden (median = 1), while the largest cluster of tumors in GC3 (48.7%, n = 4213 samples) was at the intermediate level among all three factors. figure 1C counts the number of cases in each GC for each tumor type. The color in the box represents the p value after logarithm. The test compares the sample score of a given cancer type in GC with the total sample score in the GC. red indicates enrichment, while green (negative) divergence. as shown in the figure, THCA, kirp, kirc, Pcpg, thym, and Kich, LGG are enriched in GC1. there were more ucec and coad in GC2. Stes, blca, hNSC, lusc and luad were more in GC3, OV and BRCA were higher in GC4. the rest of the cancer types were relatively evenly distributed in all GCS (Figure 1c). according to this method, the author has made a new classification of various types of cancer. In order to study the relationship between GCS and tumor immune environment, the RNA SEQ data in TCGA was analyzed to infer tumor immunity. the exploration of tumor microenvironment includes the expression of invasive immune cells (CD8 + T cells, B cells, natural killer cells and macrophages), and 70 immune related genes, including PD1 / PD-L1 From the conclusion, GC2 tumor is characterized by high level til infiltration, high expression of immune genes and up-regulated immune pathway, suggesting active or "hot" immune microenvironment. however, GC4 tumors showed low immune cell infiltration and low levels of immune gene expression, indicating no active or "cold" immune microenvironment (Figure 2a, b). this shows that there are great differences in the composition of tumor microenvironment among different gene groups. especially in the composition and activity of immune cells. this also provides theoretical support for patients with different survival rates and different responses to immunotherapy. (4) different genomic groups (GCS) are associated with patient survival. It is found that patients with different GCS have different overall survival (OS). as shown in Figure 4a, patients with tumor aggregation into GC2 (associated with thermal immune microenvironment) showed longer overall survival than patients with GC4 (associated with cold immune microenvironment), which highlights the prognostic importance of tumor immune microenvironment, regardless of cancer type. in addition, the overall survival time of GC1 patients was longer than that of GC3 patients, indicating the negative impact of scna burden on the survival of patients. cancer specific analysis also revealed the association between GCS and overall survival in a variety of cancer types, with a trend similar to that of Pan cancer analysis. 5. Different genomic groups (GCS) are associated with blocking response to immune checkpoint (ICB). In order to further study whether GCS defined by deep learning model is related to clinical effects from ICB, we applied our model to Wes data of metastatic melanoma treated with anti CTLA4 in TCGA. patients (n = 108) were divided into GC1 (n = 54), GC2 (n = 10), GC3 (n = 35) and GC4 (n = 9). GSEA analysis of 42 tumors showed that compared with GC1, GC3 had up-regulated pathways related to cell cycle, DNA repair and metabolism, while compared with GC3, GC1 had up-regulated immune related pathways, which was consistent with the results of high-quality data from TCGA resected melanoma, indicating that our model can be applied to data from small clinical samples. overall, 26 out of 108 patients achieved clinical benefits from anti CTLA4 therapy (Figure 5a). of particular interest, none of the nine patients with GC4 tumors achieved clinical benefits (Figure 5a), and their overall survival was significantly shorter than those with other GC tumors (Figure 5b). importantly, GC4 tumors are characterized by the highest scna burden, which has been reported to be associated with a lack of response to ICB and low survival. in addition, comparing patients with lower levels of scna burden (GC1 and GC2) with patients with higher levels of scnas (GC3 and GC4), patients with GC3 / GC4 tumors had significantly lower response rates (P = 0.003) and shorter overall survival (P = 0.009) than patients with GC1 / GC2 tumors. conclusion and analysis this study provides strong evidence that deep learning modeling can integrate multiple data sets, such as genome, epigenetics, transcriptomics, proteomics, etc., to discover the cross pattern correlation of multi factor input data, thus depicting the interaction between tumor and host factors. to analyze the molecular mechanism of major drug resistance in immunotherapy, to establish new predictive markers, and to accurately select patients who can benefit from immunotherapy. if you have analysis needs, you can also add a little sister wechat communication Oh, new ideas and classic ideas