Analyzing genomic and clinical data of haematological malignancies by various computational techniques

Abstract

Large-scale characterization of the human genome has enabled the extensive study of the diverse genomic alterations present in humans. The integrative analyses of the various alterations provide a detailed understanding of the factors responsible for disease initiation and its progression in disorders like cancer. There is a wide range of machine learning algorithms and statistical methods to analyze genomic data and extract information for applications such as disease diagnosis and classification of clinical subtypes. These analyses assist in developing effective drugs for specific diseases and are particularly helpful in personalized cancer therapy, where the response of a patient to a particular drug can be captured, and its correlation with the mutation profiles of the patient can be examined to design targeted medicine. Though a plethora of methods exist for analyzing cancer genomes, certain challenges exist. Therefore, in this thesis, we have formulated and proposed different computational solutions to address challenges in cancer genomics, particularly in hematological malignancies. Missing value problem is frequently observed in gene expression data, and it may significantly impact the findings extracted from the incomplete data. Therefore, we have dealt with the missing value in gene expression data by devising a compressive sensing (CS) based method, DSNN (Doubly Sparse in the Discrete Cosine Transform with Nuclear Norm minimization). A significant contribution is the utilization of Discrete Cosine Transform (DCT) based sparsity for recovering missing values. Further, we have analyzed the bulk-sequencing exome data of Multiple Myeloma (MM) and Monoclonal Gammopathy of Undetermined Significance (MGUS) patients. MM is a haematological cancer that arises from malignant transformation and deregulated proliferation of clonal plasma cells (PCs) in bone marrow, preceded by a benign condition of MGUS. The study has revealed actionable target genes that may be clinically relevant in addition to the genomic landscape of clonal evolution in MM. A statistically significant change in the mutational spectrum of MGUS and MM is observed as the disease progresses from MGUS and MM. We have also utilized survival data of the MM patients to find the association of Tumor mutational burden (TMB) with overall survival. In MM, it is critical to identify the initial risk stage of the patient as it helps in deciding the due course of the treatment to be given to the patient. Therefore, a reliable risk staging system is required, which may stratify the patients into separate subgroups and help identify patients requiring frequent visits to the hospital. Multiple staging systems have been proposed for MM, ISS and R-ISS being the gold standards used widely for MM. However, none of them uses ethnicity information. Therefore, we have developed an ethnicity-aware Artificial Intelligence (AI)-enabled risk staging system, CRSS (Consensus-based Risk Staging System), for newly diagnosed multiple myeloma patients. The proposed method can predict the risk stage of any MM patient depending on the values of the simple parameters like age, albumin, β2-microglobulin, calcium, eGFR, hemoglobin and high-risk cytogenetic information. There has been an enhanced inclination towards single-cell sequencing data over bulk-sequencing data, given the several advantages of single-cell data over bulk NGS data. However, there are different noises present in the single-cell data. Therefore, in this thesis, we have devised an optimization-based framework, ARCANE-ROG, for denoising and imputing noisy and incomplete single-cell data for inferring patterns of clonal evolution.