Computation can be performed in living cells by DNA-encoded circuits that process sensory information and control biological functions. Their construction is time-intensive, requiring manual part assembly and balancing of regulator expression. We describe a design environment, Cello, in which a user writes Verilog code that is automatically transformed into a DNA sequence. Algorithms build a circuit diagram, assign and connect gates, and simulate performance. Reliable circuit design requires the insulation of gates from genetic context, so that they function identically when used in different circuits. We used Cello to design 60 circuits forEscherichia coli(880,000 base pairs of DNA), for which each DNA sequence was built as predicted by the software with no additional tuning. Of these, 45 circuits performed correctly in every output state (up to 10 regulators and 55 parts), and across all circuits 92% of the output states functioned as predicted. Design automation simplifies the incorporation of genetic circuits into biotechnology projects that require decision-making, control, sensing, or spatial organization.

Ideal cell-free expression systems can theoretically emulate an in vivo cellular environment in a controlled in vitro platform.1 This is useful for expressing proteins and genetic circuits in a controlled manner as well as for providing a prototyping environment for synthetic biology. 2,3 To achieve the latter goal, cell-free expression systems that preserve endogenous Escherichia coli transcription-translation mechanisms are able to more accurately reflect in vivo cellular dynamics than those based on T7 RNA polymerase transcription. We describe the preparation and execution of an efficient endogenous E. coli based transcription-translation (TX-TL) cell-free expression system that can produce equivalent amounts of protein as T7-based systems at a 98% cost reduction to similar commercial systems. 4,5 The preparation of buffers and crude cell extract are described, as well as the execution of a three tube TX-TL reaction. The entire protocol takes five days to prepare and yields enough material for up to 3000 single reactions in one preparation. Once prepared, each reaction takes under 8 hr from setup to data collection and analysis. Mechanisms of regulation and transcription exogenous to E. coli, such as lac/tet repressors and T7 RNA polymerase, can be supplemented. 6 Endogenous properties, such as mRNA and DNA degradation rates, can also be adjusted. 7 The TX-TL cell-free expression system has been demonstrated for large-scale circuit assembly, exploring biological phenomena, and expression of proteins under both T7-and endogenous promoters. 6,8 Accompanying mathematical models are available. 9,10 The resulting system has unique applications in synthetic biology as a prototyping environment, or "TX-TL biomolecular breadboard."

Cell-free protein synthesis is becoming a powerful technique to construct and to study complex informational processes in vitro. Engineering synthetic gene circuits in a test tube, however, is seriously limited by the transcription repertoire of modern cell-free systems, composed of only a few bacteriophage regulatory elements. Here, we report the construction and the phenomenological characterization of synthetic gene circuits engineered with a cell-free expression toolbox that works with the seven E. coli sigma factors. The E. coli endogenous holoenzyme E(70) is used as the primary transcription machinery. Elementary circuit motifs, such as multiple stage cascades, AND gate and negative feedback loops are constructed with the six other sigma factors, two bacteriophage RNA polymerases, and a set of repressors. The circuit dynamics reveal the importance of the global mRNA turnover rate and of passive competition-induced transcriptional regulation. Cell-free reactions can be carried out over long periods of time with a small-scale dialysis reactor or in phospholipid vesicles, an artificial cell system. This toolbox is a unique platform to study complex transcription/translation-based biochemical systems in vitro.

BackgroundEscherichia coli cell-free expression systems use bacteriophage RNA polymerases, such as T7, to synthesize large amounts of recombinant proteins. These systems are used for many applications in biotechnology, such as proteomics. Recently, informational processes have been reconstituted in vitro with cell-free systems. These synthetic approaches, however, have been seriously limited by a lack of transcription modularity. The current available cell-free systems have been optimized to work with bacteriophage RNA polymerases, which put significant restrictions to engineer processes related to biological information. The development of efficient cell-free systems with broader transcription capabilities is required to study complex informational processes in vitro.ResultsIn this work, an efficient cell-free expression system that uses the endogenous E. coli RNA polymerase only and sigma factor 70 for transcription was prepared. Approximately 0.75 mg/ml of Firefly luciferase and enhanced green fluorescent protein were produced in batch mode. A plasmid was optimized with different regulatory parts to increase the expression. In addition, a new eGFP was engineered that is more translatable in cell-free systems than the original eGFP. The protein production was characterized with three different adenosine triphosphate (ATP) regeneration systems: creatine phosphate (CP), phosphoenolpyruvate (PEP), and 3-phosphoglyceric acid (3-PGA). The maximum protein production was obtained with 3-PGA. Preparation of the crude extract was streamlined to a simple routine procedure that takes 12 hours including cell culture.ConclusionsAlthough it uses the endogenous E. coli transcription machinery, this cell-free system can produce active proteins in quantities comparable to bacteriophage systems. The E. coli transcription provides much more possibilities to engineer informational processes in vitro. Many E. coli promoters/operators specific to sigma factor 70 are available that form a broad library of regulatory parts. In this work, cell-free expression is developed as a toolbox to design and to study synthetic gene circuits in vitro.

Deep neural networks with millions of parameters may suffer from poor generalization due to overfitting. To mitigate the issue, we propose a new regularization method that penalizes the predictive distribution between similar samples. In particular, we distill the predictive distribution between different samples of the same label during training. This results in regularizing the dark knowledge (i.e., the knowledge on wrong predictions) of a single network (i.e., a self-knowledge distillation) by forcing it to produce more meaningful and consistent predictions in a class-wise manner. Consequently, it mitigates overconfident predictions and reduces intra-class variations. Our experimental results on various image classification tasks demonstrate that the simple yet powerful method can significantly improve not only the generalization ability but also the calibration performance of modern convolutional neural networks.

No abstract

Lifelong learning with deep neural networks is wellknown to suffer from catastrophic forgetting: the performance on previous tasks drastically degrades when learning a new task. To alleviate this effect, we propose to leverage a large stream of unlabeled data easily obtainable in the wild. In particular, we design a novel classincremental learning scheme with (a) a new distillation loss, termed global distillation, (b) a learning strategy to avoid overfitting to the most recent task, and (c) a confidence-based sampling method to effectively leverage unlabeled external data. Our experimental results on various datasets, including CIFAR and ImageNet, demonstrate the superiority of the proposed methods over prior methods, particularly when a stream of unlabeled data is accessible: our method shows up to 15.8% higher accuracy and 46.5% less forgetting compared to the state-of-the-art method. The code is available at https://github. comB. We design a 3-step learning scheme to improve the effectiveness of global distillation: (i) training a teacher specialized for the current task, (ii) training a model by distilling the knowledge of the previous model, the teacher learned in (i), and their ensemble, and (iii) finetuning to avoid overfitting to the current task.C. We propose a confidence-based sampling method to effectively leverage a large stream of unlabeled data.

In most real-world scenarios, labeled training datasets are highly class-imbalanced, where deep neural networks suffer from generalizing to a balanced testing criterion. In this paper, we explore a novel yet simple way to alleviate this issue by augmenting less-frequent classes via translating samples (e.g., images) from more-frequent classes. This simple approach enables a classifier to learn more generalizable features of minority classes, by transferring and leveraging the diversity of the majority information. Our experimental results on a variety of class-imbalanced datasets show that the proposed method improves the generalization on minority classes significantly compared to other existing re-sampling or re-weighting methods. The performance of our method even surpasses those of previous state-of-theart methods for the imbalanced classification.