Yuanfang Guan, Ph.D. - US grants

DESCRIPTION (provided by applicant): Integrating large-scale genomics data has huge potential to accelerate the identification of disease genes in human. Three major challenges lie in the current integrative approach for predicting disease genes. First, previous integrations in general limit genomic data input to one species at a time, while disease datasets are often generated in multiple model organisms. Second, public functional genomic datasets are dominated and biased by certain data types and accessible tissues, which can be addressed by expert curation of input datasets. Third, when multiple tissue-specific networks have been generated, a mathematical formulation is lacking to prioritize among these competing networks for the specific disease under consideration. This collaborative proposal aims at addressing the above challenges by exploring a prototype of bioinformatics tools to integrate multiple relevant global and tissue-specific networks across mammalian species targeting a specific disease, here ataxia. This proposal is based on our preliminary data in developing both global and cerebellum-specific networks to prioritize ataxia associated genes, and on the two PIs' complementary expertise in genomic data integration and experimental ataxia gene confirmation. We will 1) use domain-specific and multiple species data to establish global, brain, cerebellum, related tissue, and ataxia-specific networks, and develop web tools to explore these networks; and 2) develop multiple kernel learning algorithms to weigh and integrate multiple networks to predict ataxia-associated genes. Although the algorithms will be developed targeting ataxia only, we envision that this expert-driven integrative approach will be adaptable to other disease gene identification scenarios.

Signal transduction through phosphorylation of proteins is a fundamental process involved in most of the biological pathways in all living species. It is also a dynamic process: the phosphorylation status of proteins constantly rewires in different organisms, tissues, cell lineages, and in the same cell but under different conditions. While experimental approaches such as phosphoproteomics have greatly advanced the study of such dynamics, we have only recently begun to construct computational models to estimate the phosphorylation status of proteins under unseen perturbations. The goal of this project is to develop an accurate and efficient method for predicting protein phosphorylation dynamics, and to establish an on-line service system for researchers to upload their own data and predict new dynamics under unseen perturbations. The research will implement an on-line tool, which will have wide applications from prokaryotes to higher, complex organisms, under a variety of conditions. Local high-school students will be recruited to participate in the project, which may spin off for them to compete in national and international science project competitions.

Current models for reconstructing phosphorylation relationships between proteins using time-course data mainly rely on searching the large space of possible network structures, a process that is time-consuming and not directly applicable to predicting responses under unseen interventions. This research will investigate a fundamental solution, truncated singular value decomposition with graph partitioning, to estimate phosphorylation levels given new perturbations. This method will be orders of magnitude faster and more accurate than contemporary methods. These algorithms will be tested using in silico simulation as well as phosphoproteomics data. The research will implement an on-line tool, which will allow biology users from diverse research domains to upload their time-course phosphoproteomic data, retrieve the reconstructed phosphorylation relationships specific to their organism and cell type, and predict the phosphorylation levels under unseen perturbations. The results of the project can be found at http://guanlab.ccmb.med.umich.edu/research.

PROJECT SUMMARY/ABSTRACT Finding new drugs in the seas of small molecules is a difficult task if no prior information is available. Our broad research goal is to develop innovative and accurate machine learning algorithms to predict the drug responses related to complex human diseases. Specifically, we pursue questions of how a cell line responds to a single drug and combinatorial therapies, from the perspective of biological networks and small-molecule chemoinformatics. One research goal is to understand and predict the cell line-specific responses through integrating a wide range of methods, including the propagation of drug effects via biological networks, matrix factorization of molecular profiles and chemoinformatic analysis of small molecules. We will deploy our algorithms to softwares and web servers, which will inform the downstream experimental design to identify the single and combinatorial drug candidates against human diseases. Our research program will contribute to accelerate the drug discovery process by in silico screening through large amount of potent chemicals.

People are threatened by an unprecedented pandemic: COVID-19 (caused by SARS-CoV-2). This virus is now threatening not only physical health, but also psychology, education, economy, and every corner of the infrastructure of society. So far, there is no treatment for this disease, while vaccines and neutralizing antibodies are perceived as one of the eventual solutions to this crisis. A critical piece of knowledge supporting vaccine and antibody development is understanding the genome of this virus. What is the common sequence shared among the SARS-CoV-2 strains across the globe? What are the subtypes? Are the genome variances overlapping with important genomic regions for vaccine design? Using state of the art machine learning approaches, this research will identify the shared, representative sequence across SARS-CoV-2 strains and group them by major types. This project will connect this information to the important genomic regions identified in the literature that can be used for vaccines, and thereby continuously inform the ongoing effort of vaccine development, antibody selection, and therapeutic development. The research from this study would provide society benefits through monthly updates onto web interfaces that allow the vaccine developers, the biomedical research community as well as the general public to conveniently get access to the above information. This project will support training of a graduate student in bioinformatics and provide outreach opportunities to K-12 students and the public.

This work will be a continuous effort to monthly subtype SARS-CoV-2 strains and update the shared sequences of SARS-CoV-2, in order to facilitate vaccine development and antibody design. Specifically, this research will be focusing on three objectives: 1) identifying and updating the common sequences of SARS-CoV-2 by forcing the common sequence to have the minimal evolutionary distance with all strains, or covering as many sequences as possible 2) subtyping the SARS-CoV-2 strains into major groups, which will be important to inform treatment, management and prevention measures; 3) connecting the subtyped and common genomic sequences of SARS-CoV-2 to epitopes identified in the literature. To develop vaccines or neutralizing antibody treatment, it is critical that the major variations are covered and considered. The algorithms and visualization tools will overlay the curated list of potential epitopes on top of the subtypes and the shared sequence of the virus genomes, and directly support the effort of vaccine and antibody development. This RAPID award is made by the Systematics and Biodiversity Science Cluster in the Division of Environmental Biology, using funds from the Coronavirus Aid, Relief, and Economic Security (CARES) Act.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.