Research And Implementation Of Classification For Cancer Gene Data Based On Deep Learning

With the rapid development and application of high-throughput gene sequencing technology,the cost of acquiring individual gene sequencing has been greatly reduced.It creates preconditions for researchers to research cancerat the genetic level.At present,with the advancement of information intelligence,artificial intelligence has played an important role in various fields,especially in the field of medical health.More and more researchers use deep learning to analyze gene sequencing data for predicting cancer.It can be used to assist doctors in diagnosing cancer and developing personalized treatment.Cancer research at the level of gene expression has greatly contributed to the development of cancer diagnosis and treatment technologies.This thesis uses deep learning algorithm to research and implement the classification model of cancer genome data.The main work of this thesis is as follows:1.This thesis improves a depth determination method of deep belief networks based on reconstruction error.The depth of the deep neural network is no longer entirely dependent on manual settings,but but is determined according to the reconstruction error of the RBM and the deepest network layer in the process of model training.It can reduce the randomness of manual settings to a certain extent.It enables the model to adaptively determine a better number of network layers in the pre-training process.2.This thesis proposes a cancer genomic data classification model that combines the deep belief network with LightGBM.Feature extraction of cancer gene data using a deep belief network instead of a costly artificial feature extraction process for the LightGBM classification model.The performance of the model is analyzed on the copy number variation dataset of TCGA.The experimental results show that the performance of the model can be improved by using the features extracted from the deep belief network.3.Based on the deep belief network model,this thesis proposes a classification model of breast cancer-related proteins based on multimodal data.The method extracts breast cancer subtype-related protein features contained in DNA methylation,gene expression,and miRNA expression data,respectively,through three deep belief networks.Then,a shallow neural network is established for the output of multiple single-model deep belief networks,and the back-end decision fusion is performed.The experimental results on the TCGA-BRCA dataset show that the deep learning model combining multimodal data is superior to the single mode model and the traditional classification model in multiple performance evaluation indicators.This thesis uses deep learning algorithms to research cancer genome data,focusing on cancer classification and protein status classification of breast cancer subtype.It provides a reference for the diagnosis and treatment of cancer,which can assist doctors to develop personalized treatment plans to improve the cure rate of cancer patients.