Enhanced deterministic clustering methods for cancer subtype prediction and discovery from genomic data

Abstract

The last two decades witnessed massive advancements in the technology for the extraction newline newlineof biomolecular data from tissue specimens, which resulted in the availability of high- newlinequality genomic data. The adoption of Machine Learning methods to analyze such data newline newlinehas facilitated gaining more profound insights into the molecular basis of many diseases, newlineparticularly cancer. Cancer is a heterogeneous disease. A cancer type has multiple subtypes newlinewhich differ from one another in the molecular events that trigger the disease. Machine newlineLearning methods, especially clustering algorithms, have been instrumental in predicting and newlinediscovering molecular subtypes of cancers from genomic data. They provide better results newlinethan those of the conventional methods used for the task. newlineVarious clustering methods which can be employed for the classification of cancer newlinegenomic data have been proposed in the literature. Despite this, the Biomedical research newlinecommunity has been preferring classical methods such as the Hierarchical Clustering method, newlinefor the task. Many methods lacked a considerable level of acceptance by the Biomedical newlineresearch community due to the reasons such as the inherent complexity and bias, the need newlineto assign values for a set of parameters for which finding an appropriate value is hard, and newlinethe non-deterministic nature. Accuracy is a critical aspect of cancer subtype prediction newlineand discovery . There is a pressing need for easy-to-use clustering methods having high newlineaccuracy and an ability to produce stable results. Moreover, cancer subtype discovery from newlinegenomic data is an exploratory analysis. It demands an automatic estimation of the number newlineof natural clusters in the data which is a challenging task. Therefore, it is quite relevant newlineto explore more effective and easy-to-use clustering methods that would facilitate cancer newlinesubtype prediction and discovery from genomic data. The research work carried out in this newlinedirection has resulted in three significant contributions. newline