Deep learning for population genetics

University Of Oregon, Eugene OR

Investigators

Linked publications, trials & patents

Abstract

Project Summary With the introduction of whole-genome biobank-scale data, the last half decade has seen an explosion of population genomic data but has left in its wake a gap in our ability to make sense of data of this magnitude. Whereas population genetics as a field has been traditionally data-limited, the massive volume of current sequencing means that previously unanswerable questions may now be within reach. To capitalize on this flood of information we need new methods and modes of analysis. Concurrently, the world of machine learning has been revolutionized by the rise of deep neural networks. These so-called deep learning methods offer incredible flexibility as well as astounding improvements in performance for a wide array of machine learning tasks, including computer vision, speech recognition, and natural language processing. This proposal aims to harness the great potential of deep learning for population genetic inference. In recent years our group has made great strides in pioneering the use of deep learning for population genomic analysis. The power of these methods for handling genetic data stems in part from their ability to automatically learn to extract as much useful information as possible from an alignment of DNA sequences to solve the task at hand, rather than relying on one or more predefined summary statistics which are generally problem-specific and may omit key information present in the raw data. While that is so, the deep learning methods that we and others have been developing are unable to scale to the sheer size of the data at hand, particularly whole chromosome-scale data due to intrinsic limitations on the amount of computer memory needed to process genotype matrices at this scale. In this proposal we lay out the next phase of deep learning methods that need to be developed for our computational methodology to keep pace with the glut of data being produced. In particular, we propose to design deep neural networks that can compute directly on inferred ancestral recombination graphs (ARGs) rather than genotype matrices, capitalizing on recent advances in graph learning and inference of ARGs from genetic data. We plan to develop these networks into user-friendly software tools that will be shared with the community. We will also investigate a variety of methods for estimating the uncertainty of predictions produced by deep learning methods; this area is understudied in machine learning but of great importance to biological researchers who require an accurate measure of the degree of uncertainty surrounding an estimate. Finally, we will bring self- supervised learning techniques from generative artificial intelligence that have revolutionized science and society at large, to bear in the world of population genetics. Together, these advances have the potential to transform the methodological landscape of population genetic inference.