Learning a Tracking Controller for Rolling µbots.

Micron-scale robots (µbots) have recently shown great promise for emerging medical applications. Accurate control of µbots, while critical to their successful deployment, is challenging. In this work, we consider the problem of tracking a reference trajectory using a µbot in the presence of disturbances and uncertainty. The disturbances primarily come from Brownian motion and other environmental phenomena, while the uncertainty originates from errors in the model parameters. We model the µbot as an uncertain unicycle that is controlled by a global magnetic field. To compensate for disturbances and uncertainties, we develop a nonlinear mismatch controller. We define the model mismatch error as the difference between our model's predicted velocity and the actual velocity of the µbot. We employ a Gaussian Process to learn the model mismatch error as a function of the applied control input. Then we use a least-squares minimization to select a control action that minimizes the difference between the actual velocity of the µbot and a reference velocity. We demonstrate the online performance of our joint learning and control algorithm in simulation, where our approach accurately learns the model mismatch and improves tracking performance. We also validate our approach in an experiment and show that certain error metrics are reduced by up to 40%.

Deep epistasis in human metabolism.

We extend and apply a method that we have developed for deriving high-order epistatic relationships in large biochemical networks to a published genome-scale model of human metabolism. In our analysis we compute 33,328 reaction sets whose knockout synergistically disables one or more of 43 important metabolic functions. We also design minimal knockouts that remove flux through fumarase, an enzyme that has previously been shown to play an important role in human cancer. Most of these knockout sets employ more than eight mutually buffering reactions, spanning multiple cellular compartments and metabolic subsystems. These reaction sets suggest that human metabolic pathways possess a striking degree of parallelism, inducing "deep" epistasis between diversely annotated genes. Our results prompt specific chemical and genetic perturbation follow-up experiments that could be used to query in vivo pathway redundancy. They also suggest directions for future statistical studies of epistasis in genetic variation data sets.

How Retroactivity Affects the Behavior of Incoherent Feedforward Loops.

An incoherent feedforward loop (IFFL) is a network motif known for its ability to accelerate responses and generate pulses. It remains an open question to understand the behavior of IFFLs in contexts with high levels of retroactivity, where an upstream transcription factor binds to numerous downstream binding sites. Here we study the behavior of IFFLs by simulating and comparing ODE models with different levels of retroactivity. We find that increasing retroactivity in an IFFL can increase, decrease, or keep the network's response time and pulse amplitude constant. This suggests that increasing retroactivity, traditionally considered an impediment to designing robust synthetic systems, could be exploited to improve the performance of IFFLs. In contrast, we find that increasing retroactivity in a negative autoregulated circuit can only slow the response. The ability of an IFFL to flexibly handle retroactivity may have contributed to its significant abundance in both bacterial and eukaryotic regulatory networks.

Modeling Genetic Circuit Behavior in Transiently Transfected Mammalian Cells.

Binning cells by plasmid copy number is a common practice for analyzing transient transfection data. In many kinetic models of transfected cells, protein production rates are assumed to be proportional to plasmid copy number. The validity of this assumption in transiently transfected mammalian cells is not clear; models based on this assumption appear unable to reproduce experimental flow cytometry data robustly. We hypothesize that protein saturation at high plasmid copy number is a reason previous models break down and validate our hypothesis by comparing experimental data and a stochastic chemical kinetics model. The model demonstrates that there are multiple distinct physical mechanisms that can cause saturation. On the basis of these observations, we develop a novel minimal bin-dependent ODE model that assumes different parameters for protein production in cells with low versus high numbers of plasmids. Compared to a traditional Hill-function-based model, the bin-dependent model requires only one additional parameter, but fits flow cytometry input-output data for individual modules up to twice as accurately. By composing together models of individually fit modules, we use the bin-dependent model to predict the behavior of six cascades and three feed-forward circuits. The bin-dependent models are shown to provide more accurate predictions on average than corresponding (composed) Hill-function-based models and predictions of comparable accuracy to EQuIP, while still providing a minimal ODE-based model that should be easy to integrate as a subcomponent within larger differential equation circuit models. Our analysis also demonstrates that accounting for batch effects is important in developing accurate composed models.

A formal methods approach to interpretable reinforcement learning for robotic planning.

Growing interest in reinforcement learning approaches to robotic planning and control raises concerns of predictability and safety of robot behaviors realized solely through learned control policies. In addition, formally defining reward functions for complex tasks is challenging, and faulty rewards are prone to exploitation by the learning agent. Here, we propose a formal methods approach to reinforcement learning that (i) provides a formal specification language that integrates high-level, rich, task specifications with a priori, domain-specific knowledge; (ii) makes the reward generation process easily interpretable; (iii) guides the policy generation process according to the specification; and (iv) guarantees the satisfaction of the (critical) safety component of the specification. The main ingredients of our computational framework are a predicate temporal logic specifically tailored for robotic tasks and an automaton-guided, safe reinforcement learning algorithm based on control barrier functions. Although the proposed framework is quite general, we motivate it and illustrate it experimentally for a robotic cooking task, in which two manipulators worked together to make hot dogs.

Automated Design of Pluripotent Stem Cell Self-Organization.

Human pluripotent stem cells (hPSCs) have the intrinsic ability to self-organize into complex multicellular organoids that recapitulate many aspects of tissue development. However, robustly directing morphogenesis of hPSC-derived organoids requires novel approaches to accurately control self-directed pattern formation. Here, we combined genetic engineering with computational modeling, machine learning, and mathematical pattern optimization to create a data-driven approach to control hPSC self-organization by knock down of genes previously shown to affect stem cell colony organization, CDH1 and ROCK1. Computational replication of the in vitro system in silico using an extended cellular Potts model enabled machine learning-driven optimization of parameters that yielded emergence of desired patterns. Furthermore, in vitro the predicted experimental parameters quantitatively recapitulated the in silico patterns. These results demonstrate that morphogenic dynamics can be accurately predicted through model-driven exploration of hPSC behaviors via machine learning, thereby enabling spatial control of multicellular patterning to engineer human organoids and tissues. A record of this paper's Transparent Peer Review process is included in the Supplemental Information.

Robustness analysis and tuning of synthetic gene networks.

MOTIVATION: The goal of synthetic biology is to design and construct biological systems that present a desired behavior. The construction of synthetic gene networks implementing simple functions has demonstrated the feasibility of this approach. However, the design of these networks is difficult, notably because existing techniques and tools are not adapted to deal with uncertainties on molecular concentrations and parameter values. RESULTS: We propose an approach for the analysis of a class of uncertain piecewise-multiaffine differential equation models. This modeling framework is well adapted to the experimental data currently available. Moreover, these models present interesting mathematical properties that allow the development of efficient algorithms for solving robustness analyses and tuning problems. These algorithms are implemented in the tool RoVerGeNe, and their practical applicability and biological relevance are demonstrated on the analysis of the tuning of a synthetic transcriptional cascade built in Escherichia coli. AVAILABILITY: RoVerGeNe and the transcriptional cascade model are available at http://iasi.bu.edu/%7Ebatt/rovergene/rovergene.htm.

Partial restoration of macrophage alteration from diet-induced obesity in response to Porphyromonas gingivalis infection.

Obesity is a chronic inflammatory disease that weakens macrophage innate immune response to infections. Since M1 polarization is crucial during acute infectious diseases, we hypothesized that diet-induced obesity inhibits M1 polarization of macrophages in the response to bacterial infections. Bone marrow macrophages (BMMΦ) from lean and obese mice were exposed to live Porphyromonas gingivalis (P. gingivalis) for three incubation times (1 h, 4 h and 24 h). Flow cytometry analysis revealed that the M1 polarization was inhibited after P. gingivalis exposure in BMMΦ from obese mice when compared with BMMΦ from lean counterparts. Using a computational approach in conjunction with microarray data, we identified switching genes that may differentially control the behavior of response pathways in macrophages from lean and obese mice. The two most prominent switching genes were thrombospondin 1 and arginase 1. Protein expression levels of both genes were higher in obese BMMΦ than in lean BMMΦ after exposure to P. gingivalis. Inhibition of either thrombospondin 1 or arginase 1 by specific inhibitors recovered the M1 polarization of BMMΦ from obese mice after P. gingivalis exposure. These data indicate that thrombospondin 1 and arginase 1 are important bacterial response genes, whose regulation is altered in macrophages from obese mice.

Controlling the outcome of the Toll-like receptor signaling pathways.

The Toll-Like Receptors (TLRs) are proteins involved in the immune system that increase cytokine levels when triggered. While cytokines coordinate the response to infection, they appear to be detrimental to the host when reaching too high levels. Several studies have shown that the deletion of specific TLRs was beneficial for the host, as cytokine levels were decreased consequently. It is not clear, however, how targeting other components of the TLR pathways can improve the responses to infections. We applied the concept of Minimal Cut Sets (MCS) to the ihsTLR v1.0 model of the TLR pathways to determine sets of reactions whose knockouts disrupt these pathways. We decomposed the TLR network into 34 modules and determined signatures for each MCS, i.e. the list of targeted modules. We uncovered 2,669 MCS organized in 68 signatures. Very few MCS targeted directly the TLRs, indicating that they may not be efficient targets for controlling these pathways. We mapped the species of the TLR network to genes in human and mouse, and determined more than 10,000 Essential Gene Sets (EGS). Each EGS provides genes whose deletion suppresses the network's outputs.

Exploiting the pathway structure of metabolism to reveal high-order epistasis.

BACKGROUND: Biological robustness results from redundant pathways that achieve an essential objective, e.g. the production of biomass. As a consequence, the biological roles of many genes can only be revealed through multiple knockouts that identify a set of genes as essential for a given function. The identification of such "epistatic" essential relationships between network components is critical for the understanding and eventual manipulation of robust systems-level phenotypes. RESULTS: We introduce and apply a network-based approach for genome-scale metabolic knockout design. We apply this method to uncover over 11,000 minimal knockouts for biomass production in an in silico genome-scale model of E. coli. A large majority of these "essential sets" contain 5 or more reactions, and thus represent complex epistatic relationships between components of the E. coli metabolic network. CONCLUSION: The complex minimal biomass knockouts discovered with our approach illuminate robust essential systems-level roles for reactions in the E. coli metabolic network. Unlike previous approaches, our method yields results regarding high-order epistatic relationships and is applicable at the genome-scale.

Systematic analysis of conservation relations in Escherichia coli genome-scale metabolic network reveals novel growth media.

A biochemical species is called producible in a constraints-based metabolic model if a feasible steady-state flux configuration exists that sustains its nonzero concentration during growth. Extreme semipositive conservation relations (ESCRs) are the simplest semipositive linear combinations of species concentrations that are invariant to all metabolic flux configurations. In this article, we outline a fundamental relationship between the ESCRs of a metabolic network and the producibility of a biochemical species under a nutrient media. We exploit this relationship in an algorithm that systematically enumerates all minimal nutrient sets that render an objective species weakly producible (i.e., producible in the absence of thermodynamic constraints) through a simple traversal of ESCRs. We apply our results to a recent genome scale model of Escherichia coli metabolism, in which we traverse the 51 anhydrous ESCRs of the metabolic network to determine all 928 minimal aqueous nutrient media that render biomass weakly producible. Applying irreversibility constraints, we find 287 of these 928 nutrient sets to be thermodynamically feasible. We also find that an additional 365 of these nutrient sets are thermodynamically feasible in the presence of oxygen. Since biomass producibility is commonly used as a surrogate for growth in genome scale metabolic models, our results represent testable hypotheses of alternate growth media derived from in silico analysis of the E. coli genome scale metabolic network.

Investigating metabolite essentiality through genome-scale analysis of Escherichia coli production capabilities.

MOTIVATION: A phenotype mechanism is classically derived through the study of a set of mutants and comparison of their biochemical capabilities. One method of comparing mutant capabilities is to characterize producible and knocked out metabolites. However such an effect is difficult to manually assess, especially for a large biochemical network and a complex media. Current algorithmic approaches towards analyzing metabolic networks either do not address this specific property or are computationally infeasible on the genome-scale. RESULTS: We have developed a novel genome-scale computational approach that identifies the full set of biochemical species that are knocked out from the metabolome following a gene deletion. Results from this approach are combined with data from in vivo mutant screens to examine the essentiality of metabolite production for a phenotype. This approach can also be a useful tool for metabolic network annotation validation and refinement in newly sequenced organisms. Combining an in silico genome-scale model of Escherichia coli metabolism with in vivo survival data, we uncover possible essential roles for several cell membranes, cell walls, and quinone species. We also identify specific biomass components whose production appears to be non-essential for survival, contrary to the assumptions of previous models. AVAILABILITY: Programs are available upon request from the authors in the form of Matlab script files. SUPPLEMENTARY INFORMATION: http://www.cis.upenn.edu/biocomp/manuscripts/bioinformatics_bti245/supp-info.html.