Research On Cancer Risk Prognosis Algorithm Based On Bio-omics Information

The complexity of cancer involves a series of interactions between genes and the environment.In clinical research,significant differences have been observed in the treatment efficacy and cancer development among patients with the same cancer type,which is the biggest obstacle to the development of effective cancer treatment.Prognosis refers to the estimation of the possible development and outcomes of a disease,and predicting prognosis can help doctors make more targeted diagnostic and therapeutic strategies.With the continuous development of bioinformatics and nextgeneration sequencing technologies,the main research on cancer prognosis has begun to shift towards the integration of multimodal bio-omics information.Currently,the prediction of cancer prognosis based on the combination of multiomics and multimodal data mainly focuses on using a single architecture model to process multi-omics data,with only parameter changes between models processing different types of omics data.Additionally,different types of omics data are often directly inputted into convolutional models,leading to certain shortcomings in feature representation.Lastly,ensemble learning methods such as XGBoost have shown certain advantages in processing biological data,but current research mainly focuses on using XGBoost for data dimensionality reduction or directly predicting prognosis outcomes.Based on the problems described above,the main research direction of this paper is the development of a cancer prognosis prediction model based on multi-modal and multi-omics biomarkers,aiming to help patients develop personalized and specific cancer diagnosis and treatment plans.The main contributions of this paper include:(1)The proposal of using a multi-architecture network,STCNN,to predict breast cancer prognosis using multi-omics and multimodal data,with different architecture networks used to process different types of omics data.The experiments showed that the STCNN model achieved more accurate predictions compared to using a singlearchitecture model for multi-omics data in breast cancer prognosis prediction,and the STCNN model was compared to the MDNNMD model based on the fully connected neural network,RF model,and SVM.The results showed that the STCNN model had performance improvements of 7.7%,18.8%,and 30.1% compared to the comparison models in terms of the AUC value of prognosis prediction.(2)The proposal of using fully connected layers to reorder the original cancer data distribution,making the original biological data more similar to the distribution pattern of image data and improving the prediction ability of subsequent convolutional networks.The experiments show that using fully connected layers to process prognosis prediction resulted in an AUC value improvement of 7.3%-9.3%.(3)The proposal of a two-stage prognosis prediction model,De CNNX,based on feature stacking and XGBoost.Based on the multimodal bio-omics information of hepatocellular carcinoma,lung adenocarcinoma,clear cell renal cell carcinoma,and low-grade glioma,the model is used for three-year cancer prognosis prediction.The experimental results were compared with the SWT-CNN model,RF model,and SVM model.The results showed that De CNNX was overall superior to traditional nonstacked models and other traditional deep learning and machine learning models in the above cancer prognosis results.