Md Tanbin Rahman of the Department of Biostatistics defends his dissertation on "Classification and Clustering for RNA-seq Data with Variable Selection".
Committee Chairperson: George Tseng, ScD, Department of Biostatistics
Abdus Wahed, PhD, Department of Biostatistics
Ying Ding, PhD, Department of Biostatistics
Hyun Jung Park, PhD, Department of Human Genetics
Graduate faculty of the University and all other interested parties are invited to attend
Machine learning plays an important role in genomics due to its high-dimensional structure. Classification and clustering are the two main branches of machine learning which have led to important discoveries in the field of genomics. Clustering plays crucial roles in identifying sub-types of complex disease which are not previously known while the use of classification in building predictive model is widely used in screening, prediction of the chance of developing certain diseases in the field of medicine. In transcriptomic data, where expression levels are measured on tens of thousands of genes and only a small number of subjects, the use of machine learning is often necessary. Often times, the number of genes is much higher compared to the number of samples leading to small-n-large-p problem. Variable selection in such cases is required to identify the genes that can distinguish between the different sub-types of a medical condition. In recent years, lowering of cost and high accuracy has made RNA-seq widely popular which is expected to continue to grow over the next few years. One of the important features of RNA-Seq data is that its count data structure. While there has been a great deal of literature in both clustering and classification method, most of them are either heuristic or suitable for continuous data and does not directly generalize to count data.
In Chapter 2, we propose a classifier for the count structure of the RNA-seq data with variable selection and covariate adjustment. Supervised machine learning methods have been increasingly used in biomedical research and in clinical practice. In transcriptomic applications, RNA-seq data have become dominating and have gradually replaced traditional microarray due to their reduced background noise and increased digital precision. Most existing machine learning methods are, however, designed for continuous intensities of microarray and are not suitable for RNA-seq count data. In this paper, we develop a negative binomial model via generalized linear model framework with double regularization for gene and covariate sparsity to accommodate three key elements: adequate modeling of count data with overdispersion, gene selection and adjustment for covariate effect. The proposed sparse negative binomial classifier (snbClass) is evaluated in simulations and two real applications using cervical tumor miRNA-seq data and schizophrenia post-mortem brain tissue RNA-seq data to demonstrate its superior performance in prediction accuracy and feature selection.
In Chapter 3, we will discuss a model-based clustering method which is able to use the count structure of the data without transformation thereby not losing information. Clustering with variable selection is a challenging but critical task for modern small-n-large-p data. Existing methods based on Gaussian mixture models or sparse K-means provide a solution to continuous data. With the prevalence of RNA-seq technology and lack of count data modeling for clustering, the current practice is to normalize count expression data into continuous measures and apply existing models with Gaussian assumption. In this paper, we develop a negative binomial mixture model with gene regularization to cluster samples (small n) with high-dimensional gene features (large p). EM algorithm and Bayesian information criterion are used for inference and determining tuning parameters. The method is compared with sparse Gaussian mixture model and sparse K-means using extensive simulations and two real transcriptomic applications in breast cancer and rat brain studies. The result shows superior performance of the proposed count data model in clustering accuracy, feature selection and biological interpretation by pathway enrichment analysis.
Contribution to public health:
Transcriptomic data play an important role in identifying genes that are differentially expressed under various external conditions and diseases. RNA-seq data are now the most popular method when measuring the expression level in transcriptomic data. This thesis deals with two important aspects of machine learning, namely classification and clustering for count data. The method proposed in this thesis is tailor-made for the structure of the count data produced in RNA-seq data.