Prediction Of Radiosensitivity In Breast Cancer And Head And Neck Squamous Cell Carcinoma Based On Omics Data

Breast cancer(BRCA)and head and neck squamous cell carcinoma(HNSCC)are two common malignancies with a large number of clinical patients and complete follow-up informations.At present,radiotherapy(RT)has become an important clinical treatment for these cancer patients,but not all patients derive ideal prognosis from RT due to the individual differences.Therefore,it is necessary to determine whether cancer patients are radiosensitive and can prolong survival by RT.In recent years,many studies have proposed biomarkers based on differentially expressed genes to evaluate the radiosensitivity of cancer patients.However,technical errors among different sequencing platforms will lead to differences in gene expression levels,which will lead to bias in prediction accuracy of the model.Gene pairing data processing methods may be able to eliminate this defect.In addition,considering that cancer is a metabolic disease,adding metabolism-related genes into the prediction study of cancer radiosensitivity can explore the inherent correlation between the expression level of metabolic genes and radiosensitivity,so as to more comprehensively understand the multiple influencing factors of patients’ radiosensitivity.Finally,the most prognostic models based on single omics data performed unsatisfactorily,due to the lack of the underlying systematic associations between complex molecular characteristics and cancer phenotypes.The prediction model of radiosensitivity based on multiple omics data may solve this problem.Given all that,studies on radiosensitivity prediction based on omics data are as follows:1.A radiosensitivity predicted method for BRCA patients based on metabolism-related gene signature was proposed.Firstly,we collected gene expression data and clinical data of BRCA patients with and without RT from TCGA-BRCA,E-TABM-158 and METABRIC datasets.Meanwhile,we download cancer-related metabolic gene sets from cancer cell metabolism gene database(ccm GDB).Secondly,the data was preprocessed to be gene expression value pairs.Then,the TCGA-BRCA dataset was divided into the training(N = 519)and the test(N = 519)sets,and the others were used as two independent test sets.Finally,we identified 20 prognostic related gene pairs on the training set utilizing univariate Cox proportional risk regression model,lasso regression model and multivariable Cox regression model to construct the metabolism-related genes prognostic index(MRGPI)model.The analysis showed that MRGPI model could effectively group patients,and BRCA patients in the low-index group had higher radiosensitivity,while BRCA patients in the high-index group had lower radiosensitivity.In addition,considering the inherent heterogeneity of BRCA,we further collected datasets for subtypes(Basal-like、HER2-enriched、Luminal A and Luminal B)of BRCA from three datasets(NKI,TCGA-BRCA and METABRIC).NKI dataset was used as the training set,TCGA-BRCA and METABRIC datasets as the independent test sets.Radiosensitivity signatures(RSS1,RSS2,RSS3 and RSS4)of different BRCA subtypes were constructed using Lasso-Cox on the training set.The results suggested that the radiosensitivity signature model for patients with Luminal A subtype had the best prognosis.2.A radiosensitivity prediction model for HNSCC patients based on multiple omics data was proposed.Firstly,we analyze the gene expression profiles of HNSCC patients with RT(N= 287)and without RT(N = 189)in the TCGA database by R package DESeq2,which is used for differential expression analysis,and obtained 122 differential expression genes.Secondly,287 HNSCC patients with RT were randomly divided into the training(N = 149)and the test(N = 138)sets.Then,we combined multiple omics data(gene expression,copy number variation and single nucleotide variation data)of 122 differential genes with clinical outcomes on the training set,and established a 12-gene signature scoring model by two-stage regularization and multivariable Cox regression models.Finally,using the median score of the12-gene signature on the training set as the cutoff value,the patients were divided into the highand low-score groups.We found that both on the training and test sets,the survival time of patients with RT in the low-score group was longer.It indicated that this group had higher radiosensitivity and would benefit from RT.By contrary,patients in the high-score group had lower radiosensitivity.Furthermore,we plotted a nomogram to predict the 3-and 5-year survival rates of HNSCC patients with RT.The results showed that the radiosensitivity prediction performance of 12-gene signature based on multiple omics was better than those based on single omics.We construct radiosensitivity prediction models for BRCA and HNSCC patients based on cancer omics data and clinical data,respectively,in order to screen cancer patients who could benefit from RT,which could assist doctors in estimating RT options of cancer patients.