Research On Cancer Survival Prognosis Prediction Model Based On Deep Learning

With the rise of DNA microarray technology,researchers have been provided with a large amount of high-throughput omic data.Cancer prognosis analysis and research based on biological omic data,and the formulation of corresponding treatment plans,is an effective method to improve patient survival.However,high-throughput omics data have significant characteristics such as high dimensionality,excessive noise,redundant data,and low sample size.Therefore,reducing the dimensions of the data and mining biological markers that play a decisive role in cancer prognosis are currently urgent problems to be solved,which is of great significance for exploring cancer mechanisms and improving the accuracy of cancer prognosis prediction.At present,research on prognosis prediction of breast cancer based on machine learning model has made some progress,but the feature extraction of machine learning model often involves professional knowledge in the field and the prediction effect is poor.In addition,the research on prognosis analysis based on renal cancer is less,and there is a lack of good prognosis model.To address these issues,this paper proposed two prognostic prediction models based on the BRCA and KIRC cancer transcriptome data from the TCGA database.The first model is a prognosis prediction model based on deep convolutional generation confrontation network and convolutional neural network.The positive and negative samples were divided using a three-year survival period as a threshold.First,the student t-test method and the fold-change method were used to screen differential genes to obtain a subset of differential genes,and then XGBoost was used to further screen the features of the differential gene set.After that,the two-dimensional gene image data was inputted into DCGAN for data enhancement and expansion of the dataset.Finally,the expanded dataset was inputted into a convolutional neural network for cancer prognosis prediction.The experimental results show that compared to machine learning models such as SVM,DT,and RF,the DCGAN-CNN model had a better prediction effect.At the same time,the prediction effect of the DCGAN-CNN model in two cancer gene expression data sets was improved by 3.6% and 5.7%compared to the CNN model,and the prediction performance in the BRCA dataset is the highest,with AUC is 0.87.It is worth noting that in the KIRC dataset,the predictive value of prognosis based on the machine learning model was below 0.6,while the predictive value of prognosis based on the DCGAN-CNN model reached0.73.In addition,this article discussed the impact of the fusion of mi RNA-seq dataset and lnc RNA-seq dataset on cancer prognosis prediction,and constructed a dual convolutional neural network(BCNN)model.The experimental results show that compared to the BCNN model,the DCGAN-CNN model has better overall prediction performance on both datasets.The second model is a prognosis prediction model based on short-term and short-term memory networks and convolutional neural networks.We divided positive and negative samples with a three-year survival period as the threshold value,inputted the two-dimensional gene grayscale image after differential expression analysis and XGBoost screening features into the generation antagonism neural network for data enhancement,and finally inputted the obtained dataset into a hybrid model of convolutional neural network and short-term memory network for prognosis prediction.In this paper,two prognostic models,namely the CNN-LSTM model and the LSTM-CNN model,were obtained using a hybrid approach of two different structures.The model combined the powerful feature extraction ability of CNN and the ability of LSTM to extract the correlation between genes,combining the advantages of both to achieve better prediction results.The experimental results show that the prediction effect of the LSTM-CNN model in the two cancer transcriptome datasets was generally better than that of the CNN-LSTM model.At the same time,the prediction effect of the LSTM-CNN model in the BRCA dataset was the best among all models,with an AUC value of 0.88.The prediction effect of the LSTM-CNN model in the KIRC dataset(AUC is 0.73)was comparable to that of the DCGAN-CNN model(AUC is 0.74),It can better predict the prognosis of the three-year survival time of patients with BRCA and KIRC.