Predictive Models of Boundaries of Topologically Associated Domains (TADs) in Human. The PI proposes a high-impact research project to develop novel predictive models to predict boundaries of topologically associated domains in a variety of human cell lines. The goal of this proposal is not just building these models, but to interpret them to decipher some biological meaning. Studies have recently indicated that genomes of different organisms are organized into domains called topologically associated domains (TADs), which consist of self-interacting chromatin regions. The boundaries of TADs have been linked to several diseases and gene regulation by modulating the overall genome organization. Accurate identification of these boundaries and interpretation of regulatory units in the boundaries are essential for development of effective therapy for diseases in future. Since existing experimental methods are costly and challenging, we will explore, evaluate, and compare predictive models for accurate prediction of TAD boundaries in different cell lines. We propose to test if sequence and chromatin features can be used in traditional feature based predictive models for TADs boundaries. This will be compared to deep learning models which have the capacity to ?learn? important discriminative features from sequence information only. By interpreting these models, we will characterize the regulatory code of TAD boundaries by (a) characterizing a comprehensive list of motifs in TAD boundaries in 10 different human cell lines, (b) characterizing the association between different chromatin features and their predicting power of the boundaries, and (c) scoring, ranking and prioritizing likely causal noncoding variants that occur in TAD boundaries. Furthermore, this proposal will enhance the infrastructure of research and education at University of Houston-Downtown, introducing computational biology research experiences to underrepresented minority, first time in college and female undergraduate students, who would otherwise lack such opportunities. This will allow them to acquire a broad area of skills in data analytics, critical thinking, and research methods, which will encourage them to pursue a biomedical career and change their lives and communities.
Understanding genome organization is essential for deciphering the complex mechanisms associated with gene regulation and development of effective therapies for diseases. To address current experimental challenges to quantify genome organization, we propose to develop novel artificial intelligence predictive methods to decipher genome organization in humans. These methods will also help link how genome organization modulates gene expression and how it can contribute to diseases.