Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (5 results)

  • Thesis
  • Peer Reviewed
UC Berkeley Electronic Theses and Dissertations (2024)

Genome-wide association studies (GWAS) have identified numerous genetic variants linked to various biological traits. Notably, 95% of the most likely causal variants are found in noncoding regions of the genome. To explain the effects of these noncoding variants, a promising approach is to understand how they influence phenotypes by altering gene expression. This method, known as transcriptome-wide association study (TWAS), uses genetic models of gene expression to identify genes associated with different traits. Building predictive models of gene expression from genetic variants is a crucial first step in this process. PrediXcan, a regularized linear model that sums the additive effects of genetic variants, is a popular yet older method for developing such models.

In this thesis, we explore applications and potential improvements of these models of gene expression.

In the first chapter, we enhance PrediXcan by incorporating age as an additional predictor to quantify the gene expression variance explained by both genetics and aging across 27 tissues from 948 humans. We find consistent expression heritability (h^2) among tissues, while the contribution of aging varies significantly, with R^2(age) > h^2 in five tissues. Our analysis reveals tissue-specific evolutionary trends, while aging-associated genes show patterns in proliferative tissues aligning with high cancer rates and age-associated somatic mutations. Additionally, we find that genes highly influenced by genetics tend to be less constrained and less functionally important, whereas genes that have age-associated expression are more constrained and functionally important.

In the second chapter, we evaluate an alternative approach—genomic deep learning models— for predicting gene expression from genetics. While these models perform well in predicting expression variation across genes, they fall short in predicting expression variation across individuals, not matching the performance of the simpler PrediXcan model. We also investigate the potential of combining deep learning predictions with the regularized linear regression framework of PrediXcan to enhance prediction accuracy by leveraging strengths from both approaches.

The final chapter focuses on association studies within a new, diverse cohort of people living with HIV (PLWH). We hypothesize a link between increased inflammation biomarkers activated by the NLRP3 inflammasome pathway (IL-1β, IL-6, and IL-18) and cardiovascular disease (CVD) risk within PLWH. Using whole genome sequencing data from 1001 PLWH, we identify significant GWAS signals implicating genes involved in immune function, cardiovascular function and response to HIV. We find three GTEx eQTLs matching the GWAS signal regions, suggesting a mechanism for increased CVD risk through inflammation in coronary artery tissue. Utilizing PrediXcan, we conduct a baseline transcriptome-wide association study (TWAS), providing a testbed for future model improvements, particularly in enhancing generalizability to different genetic populations, as observed in the second chapter. Together, these studies highlight the utility of gene expression models in disentangling the roles of genetics, aging, and inflammation in disease risk, and demonstrate the potential for integrating advanced modeling techniques to improve predictive accuracy and generalizability across diverse populations.

    Cover page: Interpreting the Noncoding Genome
    • Article
    • Peer Reviewed
    UC Berkeley Previously Published Works (2023)
    Genomic deep learning models can predict genome-wide epigenetic features and gene expression levels directly from DNA sequence. While current models perform well at predicting gene expression levels across genes in different cell types from the reference genome, their ability to explain expression variation between individuals due to cis-regulatory genetic variants remains largely unexplored. Here, we evaluate four state-of-the-art models on paired personal genome and transcriptome data and find limited performance when explaining variation in expression across individuals. In addition, models often fail to predict the correct direction of effect of cis-regulatory genetic variation on expression.
      Cover page: Personal transcriptome variation is poorly explained by current genomic deep learning models.
      • Article
      • Peer Reviewed
      UC Berkeley Previously Published Works (2024)
      BACKGROUND: A number of deep learning models have been developed to predict epigenetic features such as chromatin accessibility from DNA sequence. Model evaluations commonly report performance genome-wide; however, cis regulatory elements (CREs), which play critical roles in gene regulation, make up only a small fraction of the genome. Furthermore, cell type-specific CREs contain a large proportion of complex disease heritability. RESULTS: We evaluate genomic deep learning models in chromatin accessibility regions with varying degrees of cell type specificity. We assess two modeling directions in the field: general purpose models trained across thousands of outputs (cell types and epigenetic marks) and models tailored to specific tissues and tasks. We find that the accuracy of genomic deep learning models, including two state-of-the-art general purpose models-Enformer and Sei-varies across the genome and is reduced in cell type-specific accessible regions. Using accessibility models trained on cell types from specific tissues, we find that increasing model capacity to learn cell type-specific regulatory syntax-through single-task learning or high capacity multi-task models-can improve performance in cell type-specific accessible regions. We also observe that improving reference sequence predictions does not consistently improve variant effect predictions, indicating that novel strategies are needed to improve performance on variants. CONCLUSIONS: Our results provide a new perspective on the performance of genomic deep learning models, showing that performance varies across the genome and is particularly reduced in cell type-specific accessible regions. We also identify strategies to maximize performance in cell type-specific accessible regions.
        Cover page: Current genomic deep learning models display decreased performance in cell type-specific accessible regions.
        • Article
        • Peer Reviewed
        UC Berkeley Previously Published Works (2022)
        Age is the primary risk factor for many common human diseases. Here, we quantify the relative contributions of genetics and aging to gene expression patterns across 27 tissues from 948 humans. We show that the predictive power of expression quantitative trait loci is impacted by age in many tissues. Jointly modelling the contributions of age and genetics to transcript level variation we find expression heritability (h2) is consistent among tissues while the contribution of aging varies by >20-fold with [Formula: see text] in 5 tissues. We find that while the force of purifying selection is stronger on genes expressed early versus late in life (Medawar's hypothesis), several highly proliferative tissues exhibit the opposite pattern. These non-Medawarian tissues exhibit high rates of cancer and age-of-expression-associated somatic mutations. In contrast, genes under genetic control are under relaxed constraint. Together, we demonstrate the distinct roles of aging and genetics on expression phenotypes.
          Cover page: Tissue-specific impacts of aging and genetics on gene expression patterns in humans.
          • Article
          • Peer Reviewed
          UC Santa Barbara Previously Published Works (2015)
          Continuing developments in the elucidation techniques of complex catalytic processes is of foremost importance to modern synthetic chemistry, and the identification of efficient synthetic techniques relies on precise, reliable, and adaptable methods to dissect the mechanism of a given transformation. Currently, methods of reaction development are grounded upon the systematic modification of specific variables - such as temperature, time, concentration, etc. - to account for and control the dynamic series of coupled equilibria within a catalytic environment. On the other hand, tandem reaction analytical methods that involve the concomitant use of different instruments to probe a reaction can provide time-resolved information regarding active chemical species and facilitate the interrogation and optimization of the system. Herein, we report our study applying tandem in situ ReactIR and HPLC-MS monitoring to the dysprosium(III) triflate-catalyzed aza-Piancatelli rearrangement of 2-furylcarbinols, a reaction that grants access to trans-4,5-disubstituted cyclopentenones - common motifs in important biologically relevant and natural compounds. With a prototype automated sampling apparatus, information was obtained about the intrinsic chemoselectivity of the reaction, and previously unseen intermediates were observed, allowing for a more detailed reaction mechanism to be substantiated. The advantages of applying this type of tandem measurement to study these types of systems are also discussed. (Figure Presented).
            Cover page: Tandem Reaction Progress Analysis as a Means for Dissecting Catalytic Reactions: Application to the Aza-Piancatelli Rearrangement
            Top