RNA Genomics: From Classical to Modern Deep Learning

Catalog description

Data scientific analysis of genomics data has transformed biology, enabling myriad discoveries with enormous impacts on human and planetary health. Algorithms and statistics are central to knowledge of human and plant genomic variation, to microbiomes and carbon cycling in the ocean. This class will present the important open problems in the above application areas, pose them as statistical problems and explore core, unifying methods that are used to study them. We will cover diverse scientific application areas focusing on unifying ways they can be addressed statistics and informatics including (i) historical and computer-scientific approaches to addressing these problems where analysis begins with assembling and or aligning to a set of reference genomes (ii) 'statistics-first' approaches that operate on raw sequencing data to perform statistical inference for discovery. This class will present challenges and opportunities in using new methods that do not require a reference to illustrate how the planetary ecosystem can be investigated from a statistics-first perspective: from studies of microbial and plant life to humans. Motivation will be driven by current open and critical problems in planetary health, microbiome research and examples from human genomics. We will investigate statistical and informatic methods that can be used to address these problems including generalized linear models, Pearson's chi-square, permutation testing and present scientific examples/case studies where these tests fail to control the statistical level. Lectures will be pre-recorded with mandatory in-class discussions and problem sessions in class. Evaluation will be based on completion of ungraded problem sets with the major evaluation will be class projects.