Classifying DNA Methylation Imbalance Data in Cancer Risk Prediction Using SMOTE and Tomek Link Methods

Recent study shows that DNA methylation (DM) as a better biomarker and help in improving the dichotomous outcome (tumor/normal) based on several features. Over the past years, rapid advances in next-generation sequence technology had led to the timely advent of The Cancer Genome Atlas (TCGA) project which provides the most comprehensive genomic data for various kinds of Cancer. However, TCGA data is faced with the problem of class imbalance and of high data dimensionality leading to an increase in the false negative rate. In this paper, uses Synthetic Minority Oversampling Technique (SMOTE) algorithm in the pre-processing phase as a method to maintain a balanced class distribution. SMOTE is combined with the Tomek Link (T-Link) under-sampling technique for data cleaning and removing noise. To reduce the feature space of the data only those genes for which mutations have been causally implicated in cancer is considered. These are obtained through resources like Catalogue of Somatic Mutations in Cancer (COSMIC) and Clinical Interpretation of Variants in Cancer (CIViC). Classification of patient samples is then performed utilizing several machine learning algorithms of Logistic Regression, Random Forest and Gaussian Naive Bayes. Each classifier performance is evaluated using appropriate performance measures. The methodology is applied on the TCGA DNA Methylation data for 28 various cancer types, which demonstrated a superior performance in class of patient samples.

[1]  Lianghong Zheng,et al.  DNA methylation markers for diagnosis and prognosis of common cancers , 2017, Proceedings of the National Academy of Sciences.

[2]  Benjamin J. Raphael,et al.  Multiplatform Analysis of 12 Cancer Types Reveals Molecular Classification within and across Tissues of Origin , 2014, Cell.

[3]  Stephen C. Ekker,et al.  Mojo Hand, a TALEN design tool for genome editing applications , 2013, BMC Bioinformatics.

[4]  Obi L. Griffith,et al.  CIViC: A knowledgebase for expert-crowdsourcing the clinical interpretation of variants in cancer , 2016, bioRxiv.

[5]  Miron B. Kursa,et al.  Robustness of Random Forest-based gene selection methods , 2013, BMC Bioinformatics.

[6]  Jing Xia,et al.  Class Weights Random Forest Algorithm for Processing Class Imbalanced Medical Data , 2018, IEEE Access.

[7]  O. Stegle,et al.  DeepCpG: accurate prediction of single-cell DNA methylation states using deep learning , 2016, Genome Biology.

[8]  Qi Zhao,et al.  Circulating tumour DNA methylation markers for diagnosis and prognosis of hepatocellular carcinoma. , 2017, Nature materials.

[9]  D. Gifford,et al.  Predicting the impact of non-coding variants on DNA methylation , 2016, bioRxiv.

[10]  Mingming Jia,et al.  COSMIC: exploring the world's knowledge of somatic mutations in human cancer , 2014, Nucleic Acids Res..

[11]  Christian Schaefer,et al.  Homology-based inference sets the bar high for protein function prediction , 2013, BMC Bioinformatics.

[12]  Rok Blagus,et al.  SMOTE for high-dimensional class-imbalanced data , 2013, BMC Bioinformatics.

[13]  Gary J Marchionini,et al.  Information Science Roles in the Emerging Field of Data Science , 2016, J. Data Inf. Sci..

[14]  Elhassan At,et al.  Classification of Imbalance Data using Tomek Link(T-Link) Combined with Random Under-sampling (RUS) as a Data Reduction Method , 2016 .

[15]  Barbara J. Evans,et al.  Genomic Data Commons , 2017 .