Repository logo

Browsing by Author "Munsky, Brian, advisor"

Now showing 1 - 7 of 7
  • Thumbnail Image
    ItemEmbargo
    Engineering in practice: from quantitative biology modeling to engineering education
    (Colorado State University. Libraries, 2024) Weber, Lisa, author; Munsky, Brian, advisor; Atadero, Rebecca, committee member; Prasad, Ashok, committee member; Reisfeld, Brad, committee member
    In quantitative analyses of biological processes, one may use many different scales of models (e.g., spatial or non-spatial, deterministic or stochastic, time-varying or at steady-state) or many different approaches to match models to experimental data (e.g., model fitting or parameter uncertainty/ sloppiness quantification with different experiment designs). These different analyses can lead to surprisingly different results, even when applied to the same data and the same model. In Chapters 2, a variety of modeling approaches that can be utilized in analyzing biological processes are explained, with examples included of how to mathematically represent a system in order to use these various modeling approaches. Many of these mechanistic modeling approaches are demonstrated in Chapter 3 when we use a simplified gene regulation model to illustrate many of the concerns regarding modeling approach differences; these include ODE analyses of deterministic processes, chemical master equation and finite state projection analyses of heterogeneous processes, and stochastic simulations. For each analysis, we consider a time-dependent input signal (e.g., a kinase nuclear translocation) and several model hypotheses, along with simulated single cell data, to illustrate different approaches (e.g., deterministic and stochastic) in the identification of mechanisms and parameters of the same model from the same simulated data. We also explore how uncertainty in parameter space varies with respect to the chosen analysis approach or specific experiment design, and conclude with a discussion of how our simulated results relate to the integration of experimental and computational investigations to explore signal-activated gene expression models in yeast [1] and human cells [2]. Different modeling approaches are used in Chapter 4 to build on the work of Scott, et al. (2018, 2019) [3, 4] to evaluate different model classes for DNA structural conformation changes, including the unwinding/rewinding dynamics of the double-stranded DNA (dsDNA) helical structure and subsequent binding interactions with complementary single-stranded oligonucleotides probes (oligos), in relation to different conditions: temperature, salt concentration, and the level of supercoiling of the DNA molecule. This is done to identify a class of models that best fit the DNA unwinding and subsequent oligo probe binding experimental data as a function of these three conditions. In this work, we demonstrate the use of additional quantitative modeling approaches, including a modified genetic algorithm along with the process of cross validation and Markov Chain Monte Carlo (MCMC) simulations with the Metropolis-Hastings (MH) algorithm [5] to explore parameter space. We also demonstrate many of the challenges that can be encountered when modeling complex biological phenomena with actual experimental data. Although much of the work described in Chapters 2 through 4 may appear to be, on the surface, just the use of various computational methods for biological processes to increase understanding of biological mechanisms, much of it also has a separate purpose. The structure of these works and an underlying aim of much of this work, namely Chapters 2 and 3, is to provide guidance with examples to make these computational approaches more accessible to scientists and engineers. Many of these approaches are included in a quantitative biology (UQ-bio) summer school that has been conducted for the last few years as well. Through the process of developing these works and seeking to make quantitative biology more accessible, a related goal manifested to improve the accessibility of engineering education as a whole, which is addressed in Chapter 5, specifically related to diversity, equity, and inclusion (DEI) in undergraduate engineering education. There have been efforts since Fall 2017 to increase the presence of DEI in the undergraduate CBE education using a bottom up approach. To date, various efforts have been incorporated into the first two years of the CBE program. In Chapter 5, these previous efforts, along with lessons learned, are detailed. A substantial, holistic approach to incorporating DEI throughout the CBE curriculum is proposed, based on a review of recent work by other engineering education researchers, to help the CBE department create a more inclusive educational experience for undergraduate students and better enable students to handle the complex challenges they may face in their careers.
  • Thumbnail Image
    ItemOpen Access
    Exploiting noise, non-linearity, and feedback to differentially control multiple different cells using a single optogenetic input
    (Colorado State University. Libraries, 2023) May, Michael P., author; Munsky, Brian, advisor; Stasevich, Tim, advisor; Krapf, Diego, committee member; Shipman, Patrick, committee member
    Motivated by Maxwells-Demon, we propose and solve a cellular control problem in which the exploitation of stochastic noise can break symmetry between two cells and allow for specific control of multiple cells using a single input signal. We find that a new type of noise-exploiting controllers are effective and can remain effective despite coarse approximations to the model's scale or extrinsic noise in key model parameters, and that these controllers can retain performance under substantial observer-actuator time delays. We also demonstrate how SIMO controllers could drive two-cell systems to follow different trajectories with different phases and frequencies by using a noise-exploiting controller. Together, these findings suggest that noise-exploiting control should be possible even in the case where models are approximate, and where parameters are uncertain. Having demonstrated the potential of noise-enhanced feedback control through computational modeling, we have also begun the next steps toward automating microscopy to implement this potential in experimental practice. Specifically, we demonstrate a new integrated pipeline to automate the image collection including: (i) quickly search in two-dimensions to find fields of view with cells of desired phenotypes, (ii) targeted collection of three-dimensional image data for these chosen fields of view, and (iii) streamlined processing of the collected images for rapid segmentation, spot detection and tracking, and cell/spot phenotype quantification.
  • Thumbnail Image
    ItemOpen Access
    Integrating discrete stochastic models with single-cell and single-molecule experiments
    (Colorado State University. Libraries, 2019) Fox, Zachary R., author; Munsky, Brian, advisor; Stargell, Laurie, committee member; Wilson, Jesse, committee member; Prasad, Ashok, committee member
    Modern biological experiments can capture the behaviors of single biomolecules within single cells. Much like Robert Brown looking at pollen grains in water, experimentalists have noticed that individual cells that are genetically identical behave seemingly randomly in the way they carry out their most basic functions. The field of stochastic single-cell biology has been focused developing mathematical and computational tools to understand how cells try to buffer or even make use of such fluctuations, and the technologies to measure such fluctuations has vastly improved in recent years. This dissertation is focused on developing new methods to analyze modern single-cell and single-molecule biological data with discrete stochastic models of the underlying processes, such as stochastic gene expression and single-mRNA translation. The methods developed here emphasize a strong link between model and experiment to help understand, design, and eventually control biological systems at the single-cell level.
  • Thumbnail Image
    ItemOpen Access
    Machine learning methods to discover patterns in microbially driven soil carbon sequestration
    (Colorado State University. Libraries, 2020) Thompson, Jaron, author; Munsky, Brian, advisor; Metcalf, Jessica, committee member; Chan, Joshua, committee member
    Understanding how microbiomes function is a major area of research, as microorganisms play a significant role in environments spanning nearly every corner of the earth. Recent advances in DNA sequencing technology have made it possible to profile microbial communities, yet noise and sparsity in microbiome data makes it difficult to identify consistent patterns in microbial community behavior. In this thesis, we apply a host of machine learning methods to elucidate the role of the soil microbiome in mediating soil carbon sequestration. We demonstrate that broad characteristics of the soil microbiome such as richness and biomass can be used to forecast abundance of dissolved organic carbon (DOC) in soil. We also show that feature selection analysis using a host of machine learning and standard statistical techniques identifies a consensus set of significant taxa that predict DOC abundance. Finally, we demonstrate how these feature selection techniques can be used to explore more advanced probabilistic models that assign accurate estimates of prediction confidence. The methods proposed in this thesis could be used to design optimized microbial communities that combat climate change by promoting increased levels of carbon storage in soil.
  • Thumbnail Image
    ItemOpen Access
    Stochastic modeling to explore the central dogma of molecular biology and to design more informative single-molecule, live-cell fluorescence microscopy experiments
    (Colorado State University. Libraries, 2024) Raymond, William Scott, author; Munsky, Brian, advisor; Stasevich, Timothy J., committee member; Snow, Christopher D., committee member; Ben-Hur, Asa, committee member; Krapf, Diego, committee member
    Despite being described nearly a century ago, the Central Dogma of Molecular Biology still harbors many intricacies and mysteries that scientists have yet to unravel. With the convergence of many multidisciplinary scientific advances such as stronger computing power, next-generation sequencing, machine learning, and single-cell and single-molecule experiments, cellular biologists have never had more investigative power. These complex methods often are used in tandem--necessitating a closer relationship between computational biologists, computer scientists, and bench top experimentalists. As practice of this emerging dynamic, my corpus of work spans multiple areas within computational and quantitative biology with the goal to facilitate better computational tools to interpret and design experiment. For my main work at Colorado State University, I have developed the open source Python package "RNA sequence to Nascent protein simulator," rSNAPsim, to simulate Nascent Chain Tracking experiments and used it as a backbone for an entire experiment simulation pipeline to check experiment design feasibility. The rSNAPsim software provides start-to-finish capabilities for model design, model fitting, and model selection so that experimentalists can fit a mechanistic model to the Nascent Chain Tracking single-mRNA translation experiments. Along with this main work, I have provided computational modeling efforts on live-cell data on the first two steps of the Central Dogma, DNA transcription and mRNA translation. For the final entry in my corpus, I have used my interdisciplinary skills acquired at CSU to do machine learning based ncRNA riboswitch classification and discovery within the human genome; This work provides the broader scientific community with a starting point for searching for this important secondary structure within humans, where it has not been described as of time of writing.
  • Thumbnail Image
    ItemOpen Access
    Using finite state projection and Fisher information to improve single-cell experiment design to gain better understanding of DUSP1 transcription dynamics
    (Colorado State University. Libraries, 2023) Cook, Joshua A., author; Munsky, Brian, advisor; Chong, Edwin, committee member; Ghosh, Soham, committee member
    Many recent studies have combined fluorescent biochemical labels, single-cell microscopy, and discrete stochastic modeling to understand and predict how organisms react to environmental changes to control gene expression. The experimental data used in these studies is often collected using intuitively-designed applications of techniques such as single-cell immunnocytochemistry (ICC) to measure protein expression and transport or single-molecule Fluorescence in situ Hybridization (smFISH) to measure the number and position of transcribed mRNA. Once collected, these single-cell data are then analyzed using discrete stochastic models, often based on the framework of the Chemical Master Equation (CME), which can be solved using the Finite State Projection (FSP) algorithm. Unfortunately, these experiments can be expensive and labor intensive to perform, primarily due to long imaging and image analysis times, and it is not clear how these experiments must be designed to obtain the most information when their results are later analyzed using the FSP techniques. The recently discovered Finite State Projection based Fisher information Matrix (FSP-FIM) provides a potential and practical solution to this experiment design challenge by providing direct estimates for how well any potential experiment should be expected to constrain parameters for a given model or set of models. In this report, we examine this challenge of experiment design in the situation where multiple different types of experiments (i.e., ICC and smFISH) are possible, for different time points, for different numbers of measurements per time point, for different environmental inputs, and for different assumed models and combinations of unknown parameters. We extend the previous FSP-FIM theory to address these multiple challenges, and we introduce new computational tools in the form of advances to the Stochastic System Identification Toolkit (in Mathworks Matlab) that allow users to easily and efficiently compute the FSP and FIM for each of these circumstances. Using experimental smFISH data, we demonstrate the effectiveness of the FSP tools to quantitatively reproduce the single-cell transcription dynamics of the Dual Specific Phosphatase 1 (DUSP1) gene under stimulation by Dexamethasone (Dex), and we show how the FSP-FIM can be used to design optimal combinations of ICC and smFISH to further improve quantification of this gene regulatory process, including predicting the optimal allocation of measurement times to obtain the most amount of information from each experiment. To probe the generality of our results, these FSP and FSP-FIM analyses are conducted for different models, under different assumptions on known and unknown parameters, and under different drug dosage regimens. The approach developed in this work is expected to have substantial impact on how computational models can be employed to improve the selection and design of future single-cell experiments.
  • Thumbnail Image
    ItemOpen Access
    Using flow cytometry and multistage machine learning to discover label-free signatures of algal lipid accumulation
    (Colorado State University. Libraries, 2020) Tanhaemami, Mohammad, author; Munsky, Brian, advisor; Prasad, Ashok, committee member; Chitsaz, Hamidreza, committee member
    Most applications of flow cytometry or cell sorting rely on the conjugation of fluorescent dyes to specific biomarkers. However, labeled biomarkers are not always available, they can be costly, and they may disrupt natural cell behavior. Label-free quantification based upon machine-learning approaches could help correct these issues, but label replacement strategies can be very difficult to discover when applied labels or other modifications in measurements inadvertently modify intrinsic cell properties. Here we demonstrate a new, but simple approach based upon feature selection and linear regression analyses to integrate statistical information collected from both labeled and unlabeled cell populations and to identify models for accurate label-free single-cell quantification. We verify the method's accuracy to predict lipid content in algal cells Picochlorum soloecismus during a nitrogen starvation and lipid accumulation time course. Our general approach is expected to improve label-free single-cell analysis for other organisms or pathways, where biomarkers are inconvenient, expensive, or disruptive to downstream cellular processes.