Subgroup Statistical Analysis In Drug Susceptibility Testing

The understanding of disease specificity plays a key role in the development of precise medicine.We often need to properly divide patients with similar clinical manifestations into groups,so as to find a more appropriate drug for treatment of the patients.However,since there are enormous amounts of data in the human cell line,we also need to conduct feature extraction for these high-dimensional data.Therefore,this paper attempts to use a new method in the drug sensitivity test.The method can not only establish subgroup structure automatically but also evaluate the effect of specific treatment,thus exploring the relationship between cancer cell lines and drugs more deeply.For high-dimensional and small-sample data,Localized Lasso model is a kind of machine learning model that can not only establish subgroup structure but also conduct feature selection.It can use iterative least square optimization algorithm to obtain a global optimal solution and get better prediction accuracy.Based on this model,the subgroup information of cell lines is finally obtained through the analysis of cancer genomes and drug sensitivity data.And meanwhile,variable selection is carried out for each subgroup in order to get the drug sensitivity estimation value of different subgroups and further get the drug sensitivity measurement that is exclusive to each subgroup.By doing so,the two tasks,namely subgroup establishment and feature selection,can be accomplished simultaneously.Hence,precision medicine,which is based on genome information,can predict underlying diseases for individuals more accurately,so that more effective and more targeted treatment can be provided for patients in advance.Based on the Localized Lasso model,this paper analyzes 150 cell lines,the expression of 1,000 genes and the sensitivity of 24 anti-cancer drugs from the Cancer Cell Line Encyclopedia(CCLE)database.The results indicate that Localized Lasso model can not only establish subgroups among cell lines reasonably,but also select a representative gene expression for each subgroup from a large amount of gene expressions.As for nilotinib,a drug used for the treatment of ph chromosome positivechronic granulocytic leukemia,partial Lasso model divides the cell line into two categories: one is from leukemia patients and the other from non-leukemia patients.The genes MRC2 and CUEDC2,selected from the cell line of leukemia patients,have been proven to participate in the occurrence,development and drug-resistance processes of chronic granulocytic leukemia.As for AZD6244(selumetinib),a drug which has achieved good effects in the treatment of several cancers,such as colon cancer,lung cancer,breast cancer,etc.,partial Lasso model also divides the cell line into two categories and they have a significant difference.In one of the categories,50 cell lines are mostly from the tumor tissue of melanoma patients and the cell of patients with lung cancer and breast cancer.The gene ITGA3,selected from the cell line of this category,shows a significant difference between SCLC cells and NSCLC cells,and the gene MAGT1 is closely related to the growth rate of breast cancer cells.In conclusion,based on the partial Lasso model,this paper can provide patients with more accurate “targeted” therapy,and effectively improve the life quality of patients.