Optimal Configuration of Multi-Task Learning for Autonomous Driving.

For autonomous driving, it is imperative to perform various high-computation image recognition tasks with high accuracy, utilizing diverse sensors to perceive the surrounding environment. Specifically, cameras are used to perform lane detection, object detection, and segmentation, and, in the absence of lidar, tasks extend to inferring 3D information through depth estimation, 3D object detection, 3D reconstruction, and SLAM. However, accurately processing all these image recognition operations in real-time for autonomous driving under constrained hardware conditions is practically unfeasible. In this study, considering the characteristics of image recognition tasks performed by these sensors and the given hardware conditions, we investigated MTL (multi-task learning), which enables parallel execution of various image recognition tasks to maximize their processing speed, accuracy, and memory efficiency. Particularly, this study analyzes the combinations of image recognition tasks for autonomous driving and proposes the MDO (multi-task decision and optimization) algorithm, consisting of three steps, as a means for optimization. In the initial step, a MTS (multi-task set) is selected to minimize overall latency while meeting minimum accuracy requirements. Subsequently, additional training of the shared backbone and individual subnets is conducted to enhance accuracy with the predefined MTS. Finally, both the shared backbone and each subnet undergo compression while maintaining the already secured accuracy and latency performance. The experimental results indicate that integrated accuracy performance is critically important in the configuration and optimization of MTL, and this integrated accuracy is determined by the ITC (inter-task correlation). The MDO algorithm was designed to consider these characteristics and construct multi-task sets with tasks that exhibit high ITC. Furthermore, the implementation of the proposed MDO algorithm, coupled with additional SSL (semi-supervised learning) based training, resulted in a significant performance enhancement. This advancement manifested as approximately a 12% increase in object detection mAP performance, a 15% improvement in lane detection accuracy, and a 27% reduction in latency, surpassing the results of previous three-task learning techniques like YOLOP and HybridNet.

Origins of heterogeneity in Streptococcus mutans competence: interpreting an environment-sensitive signaling pathway.

Bacterial pathogens rely on chemical signaling and environmental cues to regulate disease-causing behavior in complex microenvironments. The human pathogen Streptococcus mutans employs a particularly complex signaling and sensing scheme to regulate genetic competence and other virulence behaviors in the oral biofilms it inhabits. Individual S. mutans cells make the decision to enter the competent state by integrating chemical and physical cues received from their microenvironment along with endogenously produced peptide signals. Studies at the single-cell level, using microfluidics to control the extracellular environment, provide physical insight into how the cells process these inputs to generate complex and often heterogeneous outputs. Fine changes in environmental stimuli can dramatically alter the behavior of the competence circuit. Small shifts in pH can switch the quorum sensing response on or off, while peptide-rich media appear to switch the output from a unimodal to a bimodal behavior. Therefore, depending on environmental cues, the quorum sensing circuitry can either synchronize virulence across the population, or initiate and amplify heterogeneity in that behavior. Much of this complex behavior can be understood within the framework of a quorum sensing system that can operate both as an intercellular signaling mechanism and intracellularly as a noisy bimodal switch.

Effects of Carbohydrate Source on Genetic Competence in Streptococcus mutans.

UNLABELLED: The capacity to internalize and catabolize carbohydrates is essential for dental caries pathogens to persist and cause disease. The expression of many virulence-related attributes by Streptococcus mutans, an organism strongly associated with human dental caries, is influenced by the peptide signaling pathways that control genetic competence. Here, we demonstrate a relationship between the efficiency of competence signaling and carbohydrate source. A significant increase in the activity of the promoters for comX, comS, and comYA after exposure to competence-stimulating peptide (CSP) was observed in cells growing on fructose, maltose, sucrose, or trehalose as the primary carbohydrate source, compared to cells growing on glucose. However, only cells grown in the presence of trehalose or sucrose displayed a significant increase in transformation frequency. Notably, even low concentrations of these carbohydrates in the presence of excess glucose could enhance the expression of comX, encoding a sigma factor needed for competence, and the effects on competence were dependent on the cognate sugar:phosphotransferase permease for each carbohydrate. Using green fluorescent protein (GFP) reporter fusions, we observed that growth in fructose or trehalose resulted in a greater proportion of the population activating expression of comX and comS, encoding the precursor of comX-inducing peptide (XIP), after addition of CSP, than growth in glucose. Thus, the source of carbohydrate significantly impacts the stochastic behaviors that regulate subpopulation responses to CSP, which can induce competence in S. mutans IMPORTANCE: The signaling pathways that regulate development of genetic competence in Streptococcus mutans are intimately intertwined with the pathogenic potential of the organism, impacting biofilm formation, stress tolerance, and expression of known virulence determinants. Induction of the gene for the master regulator of competence, ComX, by competence-stimulating peptide (CSP) occurs in a subpopulation of cells. Here, we show that certain carbohydrates that are common in the human diet enhance the ability of CSP to activate transcription of comX and that a subset of these carbohydrates stimulates progression to the competent state. The cognate sugar:phosphotransferase permeases for each sugar are needed for these effects. Interestingly, single-cell analysis shows that the carbohydrates that increase com gene expression do so by enhancing the proportion of cells that respond to CSP. A mathematical model is developed to explain how carbohydrates modulate bistable behavior in the system via the ComRS pathway and ComX stability.

Sharply Tuned pH Response of Genetic Competence Regulation in Streptococcus mutans: a Microfluidic Study of the Environmental Sensitivity of comX.

Genetic competence in Streptococcus mutans is a transient state that is regulated in response to multiple environmental inputs. These include extracellular pH and the concentrations of two secreted peptides, designated CSP (competence-stimulating peptide) and XIP (comX-inducing peptide). The role of environmental cues in regulating competence can be difficult to disentangle from the effects of the organism's physiological state and its chemical modification of its environment. We used microfluidics to control the extracellular environment and study the activation of the key competence gene comX. We find that the comX promoter (PcomX) responds to XIP or CSP only when the extracellular pH lies within a narrow window, about 1 pH unit wide, near pH 7. Within this pH range, CSP elicits a strong PcomX response from a subpopulation of cells, whereas outside this range the proportion of cells expressing comX declines sharply. Likewise, PcomX is most sensitive to XIP only within a narrow pH window. While previous work suggested that comX may become refractory to CSP or XIP stimulus as cells exit early exponential phase, our microfluidic data show that extracellular pH dominates in determining sensitivity to XIP and CSP. The data are most consistent with an effect of pH on the ComR/ComS system, which has direct control over transcription of comX in S. mutans.

Microfluidic study of competence regulation in Streptococcus mutans: environmental inputs modulate bimodal and unimodal expression of comX.

Streptococcus mutans regulates genetic competence through a complex network that receives inputs from a number of environmental stimuli, including two signalling peptides designated as CSP and XIP. The response of the downstream competence genes to these inputs shows evidence of stochasticity and bistability and has been difficult to interpret. We have used microfluidic, single-cell methods to study how combinations of extracellular signals shape the response of comX, an alternative sigma factor governing expression of the late competence genes. We find that the composition of the medium determines which extracellular signal (XIP or CSP) can elicit a response from comX and whether that response is unimodal or bimodal across a population of cells. In a chemically defined medium, exogenous CSP does not induce comX, whereas exogenous XIP elicits a comX response from all cells. In complex medium, exogenous XIP does not induce comX, whereas CSP elicits a bimodal comX response from the population. Interestingly, bimodal behaviour required an intact copy of comS, which encodes the precursor of XIP. The comS-dependent capability for both unimodal and bimodal response suggests that a constituent - most likely peptides - of complex medium interacts with a positive feedback loop in the competence regulatory network.

High-throughput co-culture system for analysis of spatiotemporal cell-cell signaling.

Study of spatial and temporal aspects of signaling between individual cells is essential in understanding development, the immune response, and host-pathogen interactions. We present an automated high-throughput microfluidic platform that chemically stimulates immune cells to initiate cytokine secretion, and controls the formation of signal gradients that activate neighboring cell populations. Furthermore, our system enables controlling the cell type and density based on distance, and retrieval of cells from different regions for gene expression analysis. Our device performs these tasks in 192 independent chambers to simultaneously test different co-culture conditions. We demonstrate these capabilities by creating various cellular communication scenarios between macrophages and fibroblasts in vitro. We find that spatial distribution of macrophages and heterogeneity in cytokine secretion determine spatiotemporal gene expression responses. Furthermore, we describe how gene expression dynamics depend on a cell's distance from the signaling source. Our device addresses key challenges in the study of cell-to-cell signaling, and provides high-throughput and automated analysis over a wide range of co-culture conditions.

Processing stimulus dynamics by the NF-κB network in single cells.

Cells at the site of an infection experience numerous biochemical signals that vary in amplitude, space, and time. Despite the diversity of dynamic signals produced by pathogens and sentinel cells, information-processing pathways converge on a limited number of central signaling nodes to ultimately control cellular responses. In particular, the NF-κB pathway responds to dozens of signals from pathogens and self, and plays a vital role in processing proinflammatory inputs. Studies addressing the influence of stimulus dynamics on NF-κB signaling are rare due to technical limitations with live-cell measurements. However, recent advances in microfluidics, automation, and image analysis have enabled investigations that yield high temporal resolution at the single-cell level. Here, we summarize the recent research which measures and models the NF-κB response to pulsatile and fluctuating stimulus concentrations, as well as different combinations and sequences of signaling molecules. Collectively, these studies show that the NF-κB network integrates external inflammatory signals and translates these into downstream transcriptional responses.

Discovery of New States of Immunomodulation for Vaccine Adjuvants via High Throughput Screening: Expanding Innate Responses to PRRs.

Stimulation of the innate immune system is crucial in both effective vaccinations and immunotherapies. This is often achieved through adjuvants, molecules that usually activate pattern recognition receptors (PRRs) and stimulate two innate immune signaling pathways: the nuclear factor kappa-light-chain-enhancer of activated B-cells pathway (NF-κB) and the interferon regulatory factors pathway (IRF). Here, we demonstrate the ability to alter and improve adjuvant activity via the addition of small molecule "immunomodulators". By modulating signaling activity instead of receptor binding, these molecules allow the customization of select innate responses. We demonstrate both inhibition and enhancement of the products of the NF-κB and IRF pathways by several orders of magnitude. Some modulators apply generally across many receptors, while others focus specifically on individual receptors. Modulators boost correlates of a protective immune responses in a commercial flu vaccine model and reduced correlates of reactogenicity in a typhoid vaccine model. These modulators have a range of applications: from adjuvanticity in prophylactics to enhancement of immunotherapy.

Analysis of gene expression levels in individual bacterial cells without image segmentation.

Studies of stochasticity in gene expression typically make use of fluorescent protein reporters, which permit the measurement of expression levels within individual cells by fluorescence microscopy. Analysis of such microscopy images is almost invariably based on a segmentation algorithm, where the image of a cell or cluster is analyzed mathematically to delineate individual cell boundaries. However segmentation can be ineffective for studying bacterial cells or clusters, especially at lower magnification, where outlines of individual cells are poorly resolved. Here we demonstrate an alternative method for analyzing such images without segmentation. The method employs a comparison between the pixel brightness in phase contrast vs fluorescence microscopy images. By fitting the correlation between phase contrast and fluorescence intensity to a physical model, we obtain well-defined estimates for the different levels of gene expression that are present in the cell or cluster. The method reveals the boundaries of the individual cells, even if the source images lack the resolution to show these boundaries clearly.

NF-κB memory coordinates transcriptional responses to dynamic inflammatory stimuli.

Many scenarios in cellular communication require cells to interpret multiple dynamic signals. It is unclear how exposure to inflammatory stimuli alters transcriptional responses to subsequent stimulus. Using high-throughput microfluidic live-cell analysis, we systematically profile the NF-κB response to different signal sequences in single cells. We find that NF-κB dynamics store the short-term history of received signals: depending on the prior pathogenic or cytokine signal, the NF-κB response to subsequent stimuli varies from no response to full activation. Using information theory, we reveal that these stimulus-dependent changes in the NF-κB response encode and reflect information about the identity and dose of the prior stimulus. Small-molecule inhibition, computational modeling, and gene expression profiling show that this encoding is driven by stimulus-dependent engagement of negative feedback modules. These results provide a model for how signal transduction networks process sequences of inflammatory stimuli to coordinate cellular responses in complex dynamic environments.

Spatiotemporal NF-κB dynamics encodes the position, amplitude, and duration of local immune inputs.

Infected cells communicate through secreted signaling molecules like cytokines, which carry information about pathogens. How differences in cytokine secretion affect inflammatory signaling over space and how responding cells decode information from propagating cytokines are not understood. By computationally and experimentally studying NF-κB dynamics in cocultures of signal-sending cells (macrophages) and signal-receiving cells (fibroblasts), we find that cytokine signals are transmitted by wave-like propagation of NF-κB activity and create well-defined activation zones in responding cells. NF-κB dynamics in responding cells can simultaneously encode information about cytokine dose, duration, and distance to the cytokine source. Spatially resolved transcriptional analysis reveals that responding cells transmit local cytokine information to distance-specific proinflammatory gene expression patterns, creating "gene expression zones." Despite single-cell variability, the size and duration of the signaling zone are tightly controlled by the macrophage secretion profile. Our results highlight how macrophages tune cytokine secretion to control signal transmission distance and how inflammatory signaling interprets these signals in space and time.

NF-κB responds to absolute differences in cytokine concentrations.

Cells receive a wide range of dynamic signaling inputs during immune regulation, but how gene regulatory networks measure such dynamic inputs is not well understood. Here, we used microfluidic single-cell analysis and mathematical modeling to study how the NF-κB pathway responds to immune inputs that vary over time such as increasing, decreasing, or fluctuating cytokine signals. We found that NF-κB activity responded to the absolute difference in cytokine concentration and not to the concentration itself. Our analyses revealed that negative feedback by the regulatory proteins A20 and IκBα enabled differential responses to changes in cytokine dose by providing a short-term memory of previous cytokine concentrations and by continuously resetting kinase cycling and receptor abundance. Investigation of NF-κB target gene expression showed that cells exhibited distinct transcriptional responses under different dynamic cytokine profiles. Our results demonstrate how cells use simple network motifs and transcription factor dynamics to efficiently extract information from complex signaling environments.

Bacterium in a box: sensing of quorum and environment by the LuxI/LuxR gene regulatory circuit.

The chemical signaling mechanism known as "bacterial quorum sensing" (QS) is normally interpreted as allowing bacteria to detect their own population density, in order to coordinate gene expression across a colony. However, the release of the chemical signal can also be interpreted as a means for one or a few cells to probe the local physical properties of their microenvironment. We have studied the behavior of the LuxI/LuxR QS circuit of Vibrio fischeri in tightly confining environments where individual cells detect their own released signals. We find that the lux genes become activated in these environments, although the activation onset time shows substantial cell-to-cell variability and little sensitivity to the confining volume. Our data suggest that noise in gene expression could significantly impact the utility of LuxI/LuxR as a probe of the local physical environment.

Ultra-sensitive digital quantification of proteins and mRNA in single cells.

Simultaneous measurement of proteins and mRNA in single cells enables quantitative understanding and modeling of cellular functions. Here, we present an automated microfluidic system for multi-parameter and ultra-sensitive protein/mRNA measurements in single cells. Our technology improves the sensitivity of digital proximity ligation assay by up to 55-fold, with a detection limit of 2277 proteins per cell and with detection efficiency of as few as 29 protein molecules. Our measurements using this system reveal higher mRNA/protein correlation in single mammalian cells than previous estimates. Furthermore, time-lapse imaging of herpes simplex virus 1 infected epithelial cells enabled by our device shows that expression of ICP4 -a major transcription factor regulating hundreds of viral genes- is only partially correlated with viral protein counts, suggesting that many cells go through abortive infection. These results highlight the importance of high-sensitivity protein/mRNA quantification for understanding fundamental molecular mechanisms in individual cells.

Bidirectional signaling in the competence regulatory pathway of Streptococcus mutans.

Streptococcus mutans expresses comX (also known as sigX), which encodes a sigma factor that is required for development of genetic competence, in response to the peptide signals XIP and CSP and environmental factors. XIP (sigX inducing peptide) is derived from ComS and activates comX unimodally in chemically defined media via the ComRS system. CSP (competence stimulating peptide) activates comX bimodally in peptide-rich media through the ComDE two-component system. However, CSP-ComDE activation of comX is indirect and involves ComRS. Therefore, the bimodality of CSP-dependent activation of comX may arise from either ComRS or ComDE. Here we study, at the single-cell level, how genes in the CSP signaling pathway respond to CSP, XIP and media. Our data indicate that activation of comX stimulates expression of comE. In addition, activation of comE requires intact comR and comS genes. Therefore, not only does CSP-ComDE stimulate the ComRS pathway to activate comX expression, but ComRS activation of comX also stimulates expression of the CSP-ComDE pathway and its regulon. The results demonstrate the mutual interconnection of the signaling pathways that control bacteriocin expression (ComDE) and genetic competence (ComRS), both of which are linked to lytic and virulence behaviors.