Deep learning to find prognostic-related genes in GBM patients

The deep learning survival prediction model explores genes closely related to GBM stem cells and tolerable genes that can be used to guide patient synodsCompared with the traditional Cox proportional risk survival model, deep learning can provide efficient and meaningful predictive value for patients' survival prognosis, even if there are strong clinical predictorsThese genes may be potential biomarkers or therapeutic targets- Excerpted from the article(Ref: Wong KK, et alCancers) 2019 Jan 8;11 (1)Pii: E53doi: 10.3390/cancers1010053Deep Learning-based big data analysis helps to filter prognosis-related genes from large databases, predict synergies, predict survival, etcdeep learning provides the ability to simulate the expression of a large number of different genes, is not susceptible to multiple collinearity, and better summarizes data characteristicsdeep learning based on transcriptome data has only recently been used to determine the characteristics of genes that affect GBM or other cancer prognosis."Wong KK et alof the Houston Methodist Institute and others in the GBM-related database, by studying well-trained deep learning models to discover the prognosis-related genetic characteristics of GBMarticle was published online in the January 2019 issue of The Cancersthe genes of prognosis value in patients who were first diagnosed with GBM through TCGA and underwent standardized treatment (surgery/chemotherapy)The researchers used GBM gene expression data as input data to build deep multi-layer perceptor neural network, used the like-like function as a loss function to predict the patient's survival risk, and identified the genes important to the model by input displacement methodIn addition to clinical, gene mutation and methylation factors, single-variable, multivariable Cox survival regression is also used to evaluate the predictive value of deep learning characteristics At the same time, the predictive value of deep learning to GBM was compared with other machine learning: Ridge Regression, Lasso Regression, and ElasticNet Regression Finally, 27 deep learning characteristics were extracted by deep learning to predict overall survival study first identified 39 genes from 27 models (shown in the table below) with p 0.01 as the bounding value validated in 10 glioma-related databases, and found that this group of genes significantly classified glioma patients into high-risk and low-risk groups Figure 1 Kaplan-Meier method for survival analysis, survival time (month): the first 39 genes from 7 glioblastoma and 3 low-level glioma study data will be low-risk (green), high-risk (red) patient groups, effectively separated 10 of these genes are related to GBM stem cells, stem cell microenvironments and therapeutic tolerance mechanisms: POSTN, TNR, BCAN, GAD1, TMSB15B, SCG3, PLA2G2A, NNMT, CHI3L1 and ELAVL4 deep learning survival prediction model explores genes closely related to GBM stem cells and tolerable genes that can be used to guide patient synods compared with the traditional Cox proportional risk survival model, deep learning can provide efficient and meaningful predictive value for patients' survival prognosis, even if there are strong clinical predictors these genes may be potential biomarkers or therapeutic targets.