Xiaowei Wu

Genome-wide association studies (GWAS) have successfully identified thousands of genetic loci associated with complex traits and diseases, providing critical insights into genetic architecture, biological pathways, and disease mechanisms. With the advance of machine learning, the analytical scope of GWAS can be substantially expanded by enabling joint modeling, nonlinear effects, and integrative analysis. However, deep learning approaches remain underutilized in augmenting traditional GWAS frameworks, particularly in the presence of cryptic relatedness among sampled individuals. In this paper, we propose a deep neural network (DNN)-based machine learning method to assist genetic association testing in samples with related individuals. By approximating the phenotype–genotype relationships in classical association tests and combining approximations across multiple tests, the proposed method aims to improve predictive performance in the identification of associated variants. Simulation studies demonstrate that our approach effectively complements conventional statistical methods and generally achieves increased power for detecting genetic associations. We further apply the method to data from the Framingham Heart Study, illustrating how DNN-based machine learning can facilitate the identification of genome-wide SNPs associated with average systolic blood pressure.

Venezuelan, eastern, and western equine encephalitis viruses (collectively referred to as equine encephalitis viruses---EEV) cause serious neurological diseases and pose a significant threat to the civilian population and the warfighter. Likewise, organophosphorus nerve agents (OPNA) are highly toxic chemicals that pose serious health threats of neurological deficits to both military and civilian personnel around the world. Consequently, only a select few approved research groups are permitted to study these dangerous chemical and biological warfare agents. This has created a significant gap in our scientific understanding of the mechanisms underlying neurological diseases. Valuable insights may be gleaned by drawing parallels to other extensively researched neuropathologies, such as traumatic brain injuries (TBI). By examining combined gene expression profiles, common and unique molecular characteristics may be discovered, providing new insights into medical countermeasures (MCMs) for TBI, EEV infection and OPNA neuropathologies and sequelae. In this study, we collected transcriptomic datasets for neurological disorders caused by TBI, EEV, and OPNA injury, and implemented a framework to normalize and integrate gene expression datasets derived from various platforms. Effective machine learning approaches were developed to identify critical genes that are either shared by or distinctive among the three neuropathologies. With the aid of deep neural networks, we were able to extract important association signals for accurate prediction of different neurological disorders by using integrated gene expression datasets of VEEV, OPNA, and TBI samples. Gene ontology and pathway analyses further identified neuropathologic features with specific gene product attributes and functions, shedding light on the fundamental biology of these neurological disorders. Collectively, we highlight a workflow to analyze published transcriptomic data using machine learning, which can be used for both identification of gene biomarkers that are unique to specific neurological conditions, as well as genes shared across multiple neuropathologies. These shared genes could serve as potential neuroprotective drug targets for conditions like EEV, TBI, and OPNA.

Background: With the advance of bisulfite sequencing technologies, massive amount of methylation data have been generated, which provide unprecedented opportunities to study the epigenetic mechanism and its relationship to other biological processes. A commonly seen feature of the methylation data is the correlation between nearby CpG sites. Although such a spatial correlation was utilized in several epigenetic studies, its interaction to other characteristics of the methylation data has not been fully investigated.Results: We filled this research gap from an information theoretic perspective, by exploring the impact of the spatial correlation on the methylation entropy (ME). With the spatial correlation taken into account, we derived the analytical relation between the ME and another key parameter, the methylation probability. By comparing it to the empirical relation between the two corresponding statistics, the observed ME and the mean methylation level, genomic loci under strong epigenetic control can be identified, which may serve as potential markers for cell-type specific methylation. The proposed method was validated by simulation studies, and applied to analyze a published dataset of mouse brain methylome.Conclusions: Compared to other sophisticated methods developed in literature, the proposed method provides a simple but effective way to detect CpG segments under strong epigenetic control (e.g., with bipolar methylation pattern). Findings from this study shed light on the identification of cell-type specific genes/pathways based on methylation data from a mixed cell population.

Originating from the Luria–Delbrück experiment in 1943, fluctuation analysis (FA) has been well developed to demonstrate random mutagenesis in microbial cell populations and infer mutation rates. Despite the remarkable progress in its theory and applications, FA often faces difficulties in the computation perspective, due to the lack of appropriate simulators. Existing simulation algorithms are usually designed specifically for particular scenarios, thus their applications may be largely restricted. There is a pressing need for more flexible simulators that rely on minimum model assumptions and are highly adaptable to produce data for a wide range of scenarios. In this study, we propose SimuBP, a simulator of population dynamics and mutations based on branching processes. SimuBP generates data based on a general two-type branching process, which is able to mimic the real cell proliferation and mutation process. Through simulations under traditional FA assumptions, we demonstrate that the data generated by SimuBP follow expected distributions, and exhibit high consistency with those generated by two alternative simulators. The most impressive feature of SimuBP lies in its flexibility, which enables the simulation of data analogous to real fluctuation experiments. We demonstrate the application of SimuBP through examples of estimating mutation rates.

Rapidly growing genome-wide ChIP-seq data have provided unprecedented opportunities to explore transcription factor (TF) binding under various cellular conditions. Despite the rich resources, development of analytical methods for studying the interaction among TFs in gene regulation still lags behind. In order to address cooperative TF binding and detect TF clusters with coordinative functions, we have developed novel computational methods based on clustering the sample paths of nonhomogeneous Poisson processes. Simulation studies demonstrated the capability of these methods to accurately detect TF clusters and uncover the hierarchy of TF interactions. A further application to the multiple-TF ChIP-seq data in mouse embryonic stem cells (ESCs) showed that our methods identified the cluster of core ESC regulators reported in the literature and provided new insights on functional implications of transcrisptional regulatory modules. Effective analytical tools are essential for studying protein-DNA relations. Information derived from this research will help us better understand the orchestration of transcription factors in gene regulation processes.

We propose a new approach to test associations between binary trees and covariates. In this approach, binary‐tree structured data are treated as sample paths of binary fission Markov branching processes (bMBP). We propose a generalized linear regression model and developed inference procedures for association testing, including variable selection and estimation of covariate effects. Simulation studies show that these procedures are able to accurately identify covariates that are associated with the binary tree structure by impacting the rate parameter of the bMBP. The problem of association testing on binary trees is motivated by modeling hierarchical clustering dendrograms of pixel intensities in biomedical images. By using semi‐synthetic data generated from a real brain‐tumor image, our simulation studies show that the bMBP model is able to capture the characteristics of dendrogram trees in brain‐tumor images. Our final analysis of the glioblastoma multiforme brain‐tumor data from The Cancer Imaging Archive identified multiple clinical and genetic variables that are potentially associated with brain‐tumor heterogeneity.

Dental caries is a multifactorial disease with high prevalence in both children and adults. Recent genome-wide association studies (GWASs) have revealed that genetic factors play an important role in caries incidence. However, existing methods are not sufficient to identify caries-associated genes, due to the complex correlation structure of caries GWAS data, and lack of appropriate summarization at the gene level. This paper attempts to address that by analyzing data from the Gene, Environment Association Studies (GENEVA) consortium. We investigated gene-based genetic associations for dental caries based on genome-wide data derived from the GENEVA database, with adjustment to covariates, linkage disequilibrium among single-nucleotide polymorphisms, and family relations, in sampled individuals. Several suggestive genes were identified, in which some of them have been previously found to have potential biological functions on cariogenesis. By comparing the gene sets identified from gene-based and SNP-based association testing methods, we found a non-negligible overlap, which indicates that our gene-based analysis can provide substantial supplement to the traditional GWAS analysis.

In this article, we propose a new approach for simulating trees, including their branches, sub-branches, and leaves. This approach combines the theory of biological development, mathematical models, and computer graphics, producing simulated trees and forest with full geometry. Specifically, we adopt the Lindenmayer process to simulate the branching pattern of trees and modify the available measurements and dimensions of 3D CAD developed object files to create natural looking sub-branches and leaves. Randomization has been added to the placement of all branches, sub branches and leaves. To simulate a forest, we adopt Inhomogeneous Poisson process to generate random locations of trees. Our approach can be used to create complex structured 3D virtual environment for the purpose of testing new sensors and training robotic algorithms. We look forward to applying this approach to test biosonar sensors that mimick bats' fly in the simulated environment.