Search

It has been more than a decade since the human genome was sequenced, but a complete understanding of the functional elements in the human genome is still lacking, especially for the non-coding part of the genome. The lack of complete understanding of the genome makes interpreting the function of genetic variants a daunting challenge. Here I exploited multiple ways to decipher the function of genetic variants by leveraging knowledge about transcriptional regulation and three-dimension genome organization. First, we developed SNP-SELEX, a high throughput method to assess the effect of SNPs on transcription factor (TF) binding. I demonstrated the superior performance of SNP-SELEX over previous delta PWM models, and applied results of SNP-SELEX to identify putative causal variants for type 2 diabetes. Furthermore, I employed deltaSVM algorithm to develop models that could predict the effect of SNPs on TF binding for any non-coding variants. Those models not only outperform delta PWM models in vitro and in vivo but also could help identify novel master regulator for complex traits and diseases.Next, I co-led a study to investigate the effect of genetic variants on three-dimensional (3D) chromatin conformation. I identified thousands of regions across the genome where 3D chromatin conformation varies between individuals and found those variations often accompany changes in other genome functions. Moreover, I found DNA sequence variations could influence 3D chromatin conformation and mapped hundreds of Quantitative Trait Loci (QTLs) associated with 3D chromatin features, some of which confer disease risk. Finally, I analyzed Hi-C data from human embryonic stem cells differentiated to beta cell progenitors to characterize changes in chromatin organizations during differentiation. I identified chromatin loops that are dynamic during different stages and found those loops are also associated with transcriptional regulation. Further, I revealed that chromatin loops form interaction hubs that are related to the establishment of stage-specific transcriptional programs.

The central dogma of molecular biology states that DNA is transcribed into RNA, which is then translated into proteins. The flow of genetic information in time and space is orchestrated by complex regulatory mechanisms. With the advent of modern biotechnology, our understanding of genomics, transcriptomics, and proteomics has deepened. However, bioinformatic tools for biomarker discovery in the different types of omics are still lacking. To address these issues, we developed novel algorithmic methods for three primary omics. Proteins are the main executor of cellular functions. In the proteomic level, we developed machine learning models for early diagnosis of type 2 diabetes based on the abundance of post-translational modifications (PTMs). Our models can interpret mass spectrometry data and perform integrative analysis together with clinical parameters such as HbA1C and fasting plasma glucose. In the results, we identified glycated lysine-141 of haptoglobin to be a potential biomarker. Gene regulation is conducted by cis-regulatory elements and transcription factors. In the transcriptomic level, we developed Emotif Alpha bioinformatic pipeline for DNA motif discovery and selection using RNA-seq, ChIP-seq, and gene homology data. We applied this pipeline to multiple species, including human, mouse, plants, and nematodes. The discovered motifs were validated using Gaussia Luciferase (GLuc) reporter. The 3D genome architecture in the nucleus involves spatial organization of nuclear bodies such as the histone locus body (HLB). In the 3D genomics level, we developed a bioinformatic pipeline for characterizing locus-specific chromatin interactions. Specifically, we integrated Hi-C, GAM, and SPRITE data and identified complex chromatin organization signature of the Hist1 cluster in mouse embryonic stem cell (mESC). In addition, we performed network hub analysis and identified hubs of diverse functions. These hubs contained not only histone genes and other active genes, but also lamina-associated domains (LADs) and polycomb domains. Motif and motif pair analyses further revealed putative transcription factors that might play important roles for each interaction hub.