Machine learning-based alignment-free methodology for complete genome analysis

点击翻译按钮获取中文摘要

Sequence classification is the scientific practice of identifying, naming, and grouping organisms based on differences and similarities in their genomic sequences. The problem of sequence classification is of immense importance considering that out of estimated 8.7 million (±1.3 million) species on our planet, only around 1.5 million distinct eukaryotes have been catalogued so far. This leaves us with 86% of existing species on Earth and 91% of marine species still unclassified. Due to the magnitude and complexity of the datasets, the problem of sequence comparison and analysis for the purpose of classification remains challenging.  As an Assistant Professor in Data Science, my work consists of designing and developing machine learning-based models to analyze complex data structures arising from various areas in the Natural Sciences and Engineering. I will consider two main application themes over the next five years: 1) Alignment-free genomic data analysis and 2) Model optimization using evolutionary computing. In Theme 1, I will develop an open-source, web-based, ultra-fast, and scalable machine learning-based alignment-free tool for real-time analysis and accurate classification of genomic sequences. A prototype platform will be developed to support the classification of viral genomes in the first two years. In the following two years, support for bacterial genomes and metagenomic data will be added. As another application, I will benchmark the minimum percentage of sequencing information that is a must for accurate classification. For any particular species, a subset of sequence features will be selected in a controlled manner using filtering techniques to find the minimum threshold on the length of a genomic signature (a species-specific pattern that is pervasive along the genome). As another application, I will explore the space of the genomic signature to establish a quantitative relationship between different underlying mechanisms that shape genomic signature and affect genomic integrity. In particular, the effect of environmental mutagens on the sequence composition will be studied. This may answer multiple interesting questions such as how much information in the genomic signature is because of evolution?, How much information is contributed by exposure to environmental mutagens? etc. In Theme 2, I will conduct research on optimizing the training process of machine learning models using evolutionary computing. As another application under this theme, I plan to define species-specific, quantitative genomic signature profiles based on pairwise distances. Genetic algorithms will then be applied to evolve classification rule sets based on these signature profiles and later be used to predict the labels of new unknown sequences. If successful, the use of these novel time-efficient models can then be extended to address the classification problems from other disciplines.