Cancer Grade Prediction And Pathway Analysis Based On Improved Multiple Kernel Learning Algorithm

Advances in different types of high-throughput sequencing technologies have revolutionized the researches of bioinformatics field,and make it available for researchers to generate large amounts of omics data from genomic data,to extensive transcriptomic,epigenomic,proteomic and metabolomic data very rapidly at a substantially lower cost,and what is important is that it is even increasingly common to derive omics data of distinct types from the same sample.Meanwhile,the emergence of many high quality and confidence omics databases also provides us more possibility and convenience to collect these omics data.A key goal of analysis of these data is the identification of effective models that predict phenotypic traits and outcomes,elucidating important biomarkers and generating important insights into the genetic underpinnings of the heritability of complex traits.Indeed,the emerging approaches for multiple omics data integration analysis in biological field mainly include two paradigms: multi-staged and meta-dimensional analysis.In multi-staged analysis,models are constructed using only two different scales at a time,in a stepwise,linear or hierarchical manner,when the relationship between omics data and complex trait can be described as linear or hierarchical.However,in most cases,a complex-trait is the result of a combination of omics data variation that occurs simultaneously,the multi-staged approach will fail to effectively model the complex trait.Instead,the meta-dimensional analysis would be applicable.In meta-dimensional analysis,all scales of data are combined simultaneously to identify complex,meta-dimensional models with multiple variables from different data types.Histologic grade,a complex trait of cancer,which represents the morphological assessment of tumor biological characteristics and has been shown to be able to generate important information related to the clinical behavior of cancer such as cancer's malignancy and aggressiveness,is vital in clinically planning treatment and estimating prognosis for patients.Therefore,the prediction of cancer grade can markedly elevate the detection of early cancer and efficiently guide its treatment.Although many researches have realized the significance of cancer grade,and some researches have focused on cancer grade prediction,hardly most of them tackle this problem via integrating multiple omics data.There is a need for powerful and advanced algorithm to fully harness the utility of these comprehensive high-throughput data,predicting corresponding cancer grade.In this article,we propose an improved Multiple Kernel Learning(MKL)regularized with p? norm based on meta-dimensional analysis,using the Sequential Minimal Optimization(SMO)algorithm,which is simple,easy to implement and adapt,and efficiently scales to large problems.And then,we use breast cancer as case study,and take advantage of our proposed algorithm fusing gene expression and methylation data after applying feature selection to construct a breast cancer predictor.Furthermore,we refine our model by involving biological pathway information,it can evaluate the significance of various pathways in which differential expression genes fall between different breast cancer grades.As shown in experiments results,the proposed model outperforms other state-of-the-art methods and gives us abundant biological interpretation in explaining differences among breast cancer grades.Moreover,our model can reveal an enhanced understanding of the relationship between omics data and breast cancer grades can be revealed,and provides more insights into hidden biological model of cancer.