9th Annual Southern California Regional Systems Biology Conference
Saturday, February 1st, 2020 - University of California, Riverside
Locations: Talks will be held in Life Sciences 1500, lunch will be served in the Spieth Courtyard.
Parking will be provided free of charge in Lot 6 Blue for arrivals between 7:30 a.m. and 11:30 a.m. You must obtain a parking permit from the attendant that will be stationed at the entrance of the lot from 7:30 a.m. to 11:30 a.m. on Saturday, February 1st. Pedestrian signage will be provided to direct you from Lot 6 to Life Sciences 1500 Lecture Hall for the conference.
The conference will provide an opportunity for diverse research groups from leading Southern California academic campuses to showcase their current research activities and promote collaborative interactions in all areas of Systems and Computational Biology. The conference format will consist of short faculty talks and a poster session for graduate students and postdoctoral researchers.
Mark Alber, Director, Interdisciplinary Center for Quantitative Modeling in Biology University of California Riverside - https://icqmb.ucr.edu/
Arthur Lander, Director, Center for Complex Biological Systems University of California Irvine - https://ccbs.uci.edu/
Alexander Hoffmann, Director, Institute for Quantitative and Computational Biosciences University of California Los Angeles - https://qcb.ucla.edu/
Integrating Growth Across Scales to Predict Form and Function
One of the first things a young seedling must do if push itself up through the soil to the light. This occurs through growth of the embryonic stem or ‘hypocotyl’. The hypocotyl grows primarily by cell expansion, with very few cell divisions. Strikingly, not all cells along the organ length expand at once — cell expansion starts at the organ base and moves upwards in a ‘wave’. We have been exploring 1) how cells expand more in length than width, leading to a reach for the surface; 2) how the wave of elongation might be instructed by hormonal signals; and 3) how a wave of cell elongation might result in the stereotypical organ elongation profile observed. The first two parts will be introduced followed by a simple predictive mechanical model which can recapitulate organ growth behavior from cell elongation profiles.
Engineering Next-Generation T Cells for Cancer Therapy
The adoptive transfer of T cells expressing chimeric antigen receptors (CARs) has demonstrated clinical efficacy in the treatment of advanced cancers, with anti-CD19 CAR-T cells achieving up to 90% complete remission among patients with relapsed B-cell malignancies. However, challenges such as antigen escape and immunosuppression limit the long-term efficacy of adoptive T-cell therapy. Here, I will discuss the development of next-generation T cells that can target multiple cancer antigens and resist immunosuppression, thereby increasing the robustness of therapeutic T cells against tumor defense mechanisms. Specifically, I will discuss the development of multi-input receptors and T cells that can interrogate intracellular antigens. I will also discuss the engineering of T cells that can effectively convert TGF-beta from a potent immunosuppressive cytokine into a T-cell stimulant. This presentation will highlight the potential of synthetic biology in generating novel mammalian cell systems with multifunctional outputs for therapeutic applications.
Machine Learning Prediction of Huntington’s Disease Modifier Genes
Huntington’s disease is a genetically inherited disorder, causing a progressive degeneration of the brain, involving a genetic basis resembling at least 10 other inherited neurodegenerative diseases (such as spinocerebellar ataxias), Parkinson’s and Alzheimer’s diseases. This raises the possibility that understanding the molecular basis of Huntington’s disease could provide a better understanding of a number of neurodegenerative disorders. However, the mutant protein in Huntington’s disease patients exhibits complex biophysical properties, and numerous cellular processes are simultaneously affected. Since numerous proteins interact with either the normal, the mutant, or both, proteins encoded by the Huntington’s disease gene, to decipher the features that enable this discrimination is a complex multidimensional problem. We are applying supervised and unsupervised machine learning algorithms to classify the molecular interaction partners of the Huntington’s disease protein, so as to predict genetic modifiers of the disease. These methods have been remarkably successful in discovering novel modifier genes, as confirmed by experiments with three independent model organisms. Our findings suggest a surprising molecular mechanism operative in Huntington’s disease: Chromosome condensation related DNA repair defects.
Integrating experiments of man and nature to identify new therapeutic targets and direct treatment in metabolic diseases
We live in a world of genomic investigation which promises new medicines and a more precise individual approach to treatment. Realizing this promise has been challenged by our lack of understanding of what genome scale data mean at the level of individual genes and patients. For example, hundreds of thousands of protein-altering genetic variants have been identified in sequenced populations in almost every gene in the human genome. We propose to unlock these experiments-of-nature using massively parallel bioassays to functionally characterize variants prospectively, thus enabling a genotype:function:phenotype approach. Application of the approach to insulin resistance and cardiometabolic diseases will be presented.
A coupled model of tissue mechanics and morphogen transport predicts the mechanism of lung branching morphogenesis
This talk presents our recent findings on lung branching morphogenesis. Lung branching morphogenesis is a complex and elegant process that gives the lungs its highly stereotyped fractal structure. It begins during the embryonic stage of development and is regulated primarily by morphogens and biomechanical factors. It remains unknown how these key factors interact to regulate the branching process, thus the motivation of this research. We build mathematical models to study the regulation of embryonic lung branching. We develop a partial differential equation model to study the production and diffusion of key morphogens in the tissue microenvironment. We model the biomechanical forces using an elastic model. We couple these two different models and investigate how morphogens and biomechanical forces interact during lung branching morphogenesis. Simulation of the model is done using the finite element method on a realistic geometry that represents the embryonic lungs. In this talk, I will discuss new findings that show that the geometry of the tissue is a key parameter that links morphogen dynamics to biomechanical factors. In particular, I will show that the biomechanical forces that act on the lungs, modify the geometry of the lung tissue thereby controlling the flux distribution of morphogens, the diameter of branches and the positional information for branching. These results agree with laboratory experiments and predict mechanisms of lung morphogenesis that could be tested experimentally.
Imaging and Analysis of Emergent Dynamics in a Developing Nervous System
All animals are born with a set of reflexes and behaviors that enable them to survive after birth, yet little is known about the mechanisms by which the neural circuitry that defines and coordinates these behaviors are established and refined during embryonic development. While the earliest stages of synapse formation are thought to be independent of neural activity, this eventually gives way to a plastic nervous system capable of optimizing behavioral output and refining its connectivity. We aim to understand the mechanisms that shape and refine early neural circuits during embryogenesis through studies of the nematode C. elegans. Prior decades of work have produced a wealth of knowledge regarding the genetic specification of its nervous system and the complete chemical and electrical connectivity of the 300 neurons in the C. elegans brain. With this detailed understanding of the post-embryonic brain, we are using genetic and optical perturbations to dissect the function of neural circuits forming during this critical period of development. We are also actively developing image analysis and computational methodologies to describe emergent neural and behavioral dynamics in the C. elegans embryo with the ultimate goal of understanding the developmental basis of neural circuit formation and maturation.
Mathematical Modeling of Iron Metabolism
Computational and mathematical modeling has become an important tool for modern life-sciences research. Understanding the main molecular players, interactions, and metabolic pathways of iron metabolism is one of the goals of this work. We approach this problem by building mathematical models from a dynamical systems point of view and by using experimental data to test, confirm, and predict new models. Our preliminary findings have started to elucidate some of the main pathways associated in iron regulation. In this talk, I will give a brief overview of iron metabolism and explain some of our experimental and computational findings.
To live or die: cell decision-making in the face of viral infection
Paradoxically, TNFa simultaneously activates opposing pro-apoptotic and pro-survival signaling. We show that the activation of antagonistic pathways changes the properties of cell fate decision transitioning cells from a “slow and accurate” to “fast and error-prone” decision mode. Mathematical modeling predicts, and experiments in vitro and in vivo validate, that the regulation of the decision mode of non-immune cells by macrophages is key to the prevention of viral spread. Overall our results demonstrate how a collective phenotype emerges from the regulation of fundamental tradeoffs within cellular cognitive processes.
Metabolomics and Stable Isotope Tracing to Understand Aging and to Design Combinatorial Cancer Drugs
Metabolism plays a central role in regulating biology. Thanks to advances in mass spectrometry, metabolomics provides a systems level understanding of the cellular metabolic state. This includes measurement of both metabolite concentrations (pool size) and pathway flux through stable isotope tracing. Here, I will discuss how pool size and pathway flux reveal different but complementary information and then discuss the application of metabolomics to understand replicative senescence in primary epithelial cells and to design drug combinations that exploit the metabolic vulnerabilities of cancer cells. Together, these vignettes will demonstrate the power of metabolomics to elucidate the regulatory role of metabolism in biology.
What's in a SMS genome?
In the past 12 months, third-generation, single-molecule sequencing (SMS) instruments have increased throughput to become palatable for small studies in human genetics. More importantly, the technologies are now within reach of powering large-scale studies using SMS. This begs the question: how will our understanding of human genetics change with a new approach to sequencing genomes? In this talk I will describe the successes and challenges of using SMS to detect variation in human genomes. SMS may be used to produce haplotype-resolved de novo assemblies and detect base-pair resolved variation. Relative to short-read approaches, this detects roughly a six-fold increase in structural variants: insertions, deletions, and rearrangements at least 50 bases. However, there remains challenges for translating this improved sensitivity into an improved diagnostic yield. In three healthy individuals sequenced by both SMS and short-reads, most variants affecting genes known to be intolerant to loss of function were detected by both approaches, similarly for over 85% of variants affecting functional non-coding DNA. We will conclude by discussing new approaches we are developing to resolve variation currently missed by SMS, and genotyping methods to relate newly discovered variation from SMS genomes to large cohorts sequenced by short-read studies.
Towards Multiscale Mechanical Characterization of the Lung
Unbeknownst to most, lung disease tops the rankings as the leading cause of death in the world. New approaches to understand lung biomechanics are becoming increasingly important in order to be able to investigate how disease impacts pulmonary function. In this talk, we will share our lab's new advances to experimental capabilities, spanning organ to tissue scale, in order to inform computational techniques. Results have implications for clinical ventilation and understanding adaptation and remodeling in diseased states.
Combining Data, Control Theory, Statistical Thermodynamics with Machine Learning to Predict Enzyme Regulation, Metabolite Concentrations and Rate Constants
1 Biological Sciences Division, Pacific Northwest National Laboratory
2 Department of Mathematics, University of California, Riverside, Riverside, CA 92521
Experimental measurement or computational inference/prediction of the enzyme regulation needed in a metabolic pathway is hard problem. Consequently, regulation is known only for well-studied reactions of central metabolism in a few organisms. In this study, we use statistical thermodynamics and metabolic control theory as a theoretical framework to determine the enzyme activities that are needed to control metabolite concentrations such that they are consistent with experimentally measured values. A reinforcement learning approach is utilized to learn optimal regulation policies that match physiological levels of metabolites while maximizing the entropy production rate and minimizing the heat loss. The learning takes a minimal amount of time, and efficient regulation schemes were learned that agree surprisingly well with known regulation. The learning is facilitated by a new approach in which steady state solutions are obtained by convex optimization rather than ODE solvers, making the time to solution seconds rather than days. The optimization is based on the Marcelin-De Donder formulation of mass action kinetics, from which rate constants are inferred. Consequently, a full ODE-based, mass action simulation with rate parameters and post-translational regulation is obtained. We demonstrate the process on three pathways in the central metabolism E. coli (gluconeogenesis, glycolysis-TCA, Pentose Phosphate-TCA) that each require different regulation schemes.
Combining Data, Control Theory, Statistical Thermodynamics with Machine Learning to Predict Enzyme Regulation, Metabolite Concentrations and Rate Constants
Prion proteins are responsible for a variety of neurodegenerative diseases in mammals such as Creutzfeldt-Jakob disease in humans and "mad-cow" disease in cattle. While these diseases are fatal to mammals, a host of harmless phenotypes have been associated with prion proteins in S. cerevisiae, making yeast an ideal model organism for prion diseases. Most mathematical approaches to modeling prion dynamics have focused on either the protein dynamics in isolation, absent from a changing cellular environment, or modeling prion dynamics in a population of cells by considering the "average" behavior. However, such models have been unable to recapitulate in vivo properties of yeast prion strains including experimentally observed rates of prion loss.
My group develops physiologically relevant mathematical models by considering both the prion aggregates and their yeast host. We then validate our model and infer parameters through carefully designed in vivo experiments. In this talk, I will present two recent results. First, we adapt the nucleated polymerization model for aggregate dynamics to a stochastic context to consider a rate limiting event in the establishment of prion disease: the rst the successful amplication of an aggregate. We then develop a multi-scale aggregate and generation structured population model to study the amplication of prion aggregates in a growing population of cells. In both cases, we gain new insights into prion phenotypes in yeast and quantify how common experimentally observed outcomes depend on population heterogeneity.
Using Operator Theoretic Methods in Biological Systems to Debug Circuits and Extract Sensors for Cell-State
In this talk I will discuss data-driven versus reductionist approaches to discover biological dynamics from time series measurements data. I briefly review the method of Koopman, developed in the last two decades by Mezic, Rowley, Kutz, Brunton, and many others, as a means for data-driven inference. I describe our recent work to integrate deep learning and dynamic mode decomposition, an algorithm for discovering Koopman operators and illustrate it's utility on several simulated and experimental datasets. Motivated by these successes, I present the challenge of estimating Koopman operators using sparsely sampled time-series RNAseq data and show how learning the Koopman operator for such datasets requires a robust optimization framework, which yields mathematically nice insight into regularization problems. We then consider a practical variant of RNAseq modeling, where noise levels vary from gene to gene and show extensions of a sparse-robust DMD algorithm for treating heterogeneous noise. I illustrate empirically how this is often the case in biological data and emphasize how biological replicates (or reproducibility) bootstrap the learning problem. For the remainder of the talk, I direct our attention on recent reductionist biological experiments performed in collaboration with MIT, Ginkgo Bioworks, and the UT Austin Texas Advanced Computing Center. We show how the structure of the experiments (and the so-called modularity principle) lends a natural adaptation of dynamic mode decomposition algorithms to what are essentially structured system identification problems. We show how the reductionist structure of biological experiments naturally ameliorates the complexity of closed-loop identification problems in biology, while a Koopman approach enables affine separation and hierarchical learning of the system's dynamics in layers. Our case study centers on a NAND gate implemented as a biological circuit in E. coli, consisting of 29 engineered biological parts and over 7000 basepairs of synthetic genetic code. As we apply the structured DMD algorithm on transcriptomics (RNAseq) data, we discover an interface between natural host genes of E. coli and the dynamics of the circuit that is astonishingly highly connected. Each genetic part has a different footprint, or impact, on the natural host's dynamics! Of 429 analyzed genes, we discover 420 genes are significantly up- or down-regulated due to the introduction of one synthetic gene, while another synthetic gene only impacts 269 genes. The combination of data-driven learning algorithms with model-guided biological experiments gives rise to a novel perspective on host-circuit impact. While previous measures of circuit impact quantify impact in terms of reduction of biological fitness, we show that a genetic circuit can have a significant impact on a biological host, even in the absence of biological fitness! Finally, our analysis reveals that the rewired host-circuit interface ultimately alters the network topology of our NAND gate, suggesting a new strategy for genetic circuit design.
Towards Understanding Cell Migration by Synthesizing Machine Learning and Mathematical Modeling
How and why cells move, collectively and singly, is still poorly understood. Mathematical modeling approaches to answering this question, such as ordinary or partial differential equations, are interpretable and are able to predict future states. However these models require constant refinement and are laborious to develop. On the other hand, machine learning approaches are quick, but may lack interpretability and the ability to predict. Here we combine mathematical modeling approaches with machine learning methods to obtain accurate, interpretable models directly from spatiotemporal data, bypassing the need for model refinement. We test the hybrid methodology with examples from biological applications such as wound healing, cell movement, and cancer.
Mechanics of Simulated Membranes: Extracting Bending Moduli and Related Elastic Properties
Detailed molecular simulations are increasingly used in membrane biophysics to assist in the interpretation of experiments. However, many of the most fundamental physical properties prove difficult to accurately measure in silico due to small system sizes. The membrane bending modulus is one well-known example of this problem. A computational strategy will be presented to accurately extract membrane bending moduli (lipid tilt and twist moduli too) directly from fully atomistic simulations.
Epigenetic Effects of Transposable Elements in 3D Nuclear Space Impact Genome Evolution
Transposable elements (TEs) are ubiquitous genome parasites whose evolution is tightly intertwined with the function and evolution of host genomes. They are abundant in pericentromeric heterochromatin (PCH), which is enriched for the repressive epigenetic mark H3K9me2/3 and its reader protein, HP1a. TEs are also prevalent in the gene-rich euchromatic genome and, interestingly, can lead to tens of Kb enrichment of H3K9me2/3 at flanking euchromatic sequences. Due to the biophysical properties of HP1a, PCH of different chromosomes can coalesce into a single domain within the 3D nuclear space through liquid-liquid fusion (PCH domain). This domain is enriched with silencing proteins and can significantly influence the function of genes that are recruited/brought into this phase-separated 3D space. We hypothesized that euchromatic TEs enriched for H3K9me2/3 and HP1a could also spatially interact with the main PCH domain, influencing euchromatic genome function. To investigate the spatial contacts of euchromatic loci with PCH, we developed a novel analysis method that incorporates Hi-C reads originating from repetitive PCH DNAs, which were excluded from previous Hi-C studies. Despite being far from PCH on a linear chromosome, ~14% euchromatic TEs show 3D interactions with PCH, which were validated with locus-specific FISH using Oligopaint. We leveraged polymorphic (presence/absence) TE insertions in natural populations and compared the 3D organization of homologous sequences with and without TE-induced H3K9me2/3 enrichment, which showed that spatial contacts between euchromatic loci and PCH require the presence of repressive marks. Importantly, population genetic analysis revealed that TEs spatially interacting with PCH are more strongly selected against, suggesting functional consequence of these 3D contacts with PCH. Our findings demonstrate that naturally occurring TEs could significantly influence the 3D organization of the genome, having a far-reaching impact on the function and evolution of the gene-rich euchromatic genome.
Structure and Dynamics of Bacterial Collectives
The structure and dynamics of populations are affected by the way they respond to stress. In particular, populations may employ evolutionary mechanisms that favor the survival of the collective. We investigate the impact of stress from bacteriophage (viral) infection and antibiotics on the structure and dynamics of collectives of the bacterium Pseudomonas aeruginosa. We find that phage-infected bacterial sub-populations isolate themselves from uninfected (healthy) sub-populations by releasing a stress signal. This mechanism has the overall effect of limiting the infection to a sub-population, which promotes the survival of the overall population. Antibiotics also cause the release of the stress signal, which repels untreated bacteria from approaching the area containing antibiotics, enabling sub-populations to evade antibiotic treatment altogether. The stress responses observed here could increase bacterial resilience against antibiotic treatment and phage therapy in healthcare settings, as well as provide a simple evolutionary strategy to avoid areas containing stress.