Over the past two or three years, the pace of digital transformation is increasing thanks to the improved performance, power, and adaptability of tools, and investments in cloud computing, data architecture, and visualization technologies. There are also an increasing number of use cases for machine learning and, in future, quantum computing, which will accelerate the development of molecules and formulations. The broad digital transformation taking place in R&D is allowing researchers to automate time-consuming manual processes and opening new research horizons in thorny problems that have failed to elicit breakthroughs. This new report, based on interviews with R&D executives at companies including Novartis, Roche, Merck, Syngenta, and BASF, explores the use cases, best practices, and roadmaps for digitalizing science. Rich, accessible, and shareable data are the fuel on which today's breakthrough analytics and computing tools rely.

This paper introduces BioScript, a domain-specific language (DSL) for programmable biochemistry that executes on emerging microfluidic platforms. The goal of this research is to provide a simple, intuitive, and type-safe DSL that is accessible to life science practitioners. The novel feature of the language is its syntax, which aims to optimize human readability; the technical contribution of the paper is the BioScript type system. The type system ensures that certain types of errors, specific to biochemistry, do not occur, such as the interaction of chemicals that may be unsafe. Results are obtained using a custom-built compiler that implements the BioScript language and type system. The last two decades have witnessed the emergence of software-programmable laboratory-on-a-chip (pLoC) technology, enabled by technological advances in microfabrication and coupled with scientific understanding of microfluidics, the fundamental science of fluid behavior at the micro- to nanoliter scale. The net result of these collective advancements is that many experimental laboratory procedures have been miniaturized, accelerated, and automated, similar in principle to how the world's earliest computers automated tedious mathematical calculations that were previously performed by hand. Although the vast majority of microfluidic devices are effectively application-specific integrated circuits (ASICs), a variety of programmable LoCs have been demonstrated.16, With a handful of exceptions, research on programming languages and compiler design for programmable LoCs has lagged behind their silicon counterparts. To address this need, this paper presents a domain-specific programming language (DSL) and type system for a specific class of pLoC that manipulate discrete droplets of liquid on a two-dimensional grid. The basic principles of the language and type system readily generalize to programmable LoCs, realized across a wide variety of microfluidic technologies.

Graph representation learning has attracted increasing research attention. However, most existing studies fuse all structural features and node attributes to provide an overarching view of graphs, neglecting finer substructures' semantics, and suffering from interpretation enigmas. This paper presents a novel hierarchical subgraph-level selection and embedding based graph neural network for graph classification, namely SUGAR, to learn more discriminative subgraph representations and respond in an explanatory way. SUGAR reconstructs a sketched graph by extracting striking subgraphs as the representative part of the original graph to reveal subgraph-level patterns. To adaptively select striking subgraphs without prior knowledge, we develop a reinforcement pooling mechanism, which improves the generalization ability of the model. To differentiate subgraph representations among graphs, we present a self-supervised mutual information mechanism to encourage subgraph embedding to be mindful of the global graph structural properties by maximizing their mutual information. Extensive experiments on six typical bioinformatics datasets demonstrate a significant and consistent improvement in model quality with competitive performance and interpretability.

In this episode, Abate follows up with Benjamin Pietro Filardo, founder of Pliant Energy Systems and NACROM, the North American Consortium for Responsible Ocean Mining. Pietro discusses the current proposed solutions for deep sea mining which are environmentally destructive, and he offers an alternative solution using swarm robots which could mine the depths of the ocean while creating minimal disturbance to this mysterious habitat. Benjamin "Pietro" Filardo After several years in the architectural profession, Pietro founded Pliant Energy Systems to explore renewable energy concepts he first pondered while earning his first degree in marine biology and oceanography. With funding from four federal agencies he has broadened the application of these concepts into marine propulsion and a highly novel robotics platform.

Visual steel surface defect detection is an essential step in steel sheet manufacturing. Several machine learning-based automated visual inspection (AVI) methods have been studied in recent years. However, most steel manufacturing industries still use manual visual inspection due to training time and inaccuracies involved with AVI methods. Automatic steel defect detection methods could be useful in less expensive and faster quality control and feedback. But preparing the annotated training data for segmentation and classification could be a costly process. In this work, we propose to use the Transfer Learning-based U-Net (TLU-Net) framework for steel surface defect detection. We use a U-Net architecture as the base and explore two kinds of encoders: ResNet and DenseNet. We compare these nets' performance using random initialization and the pre-trained networks trained using the ImageNet data set. The experiments are performed using Severstal data. The results demonstrate that the transfer learning performs 5% (absolute) better than that of the random initialization in defect classification. We found that the transfer learning performs 26% (relative) better than that of the random initialization in defect segmentation. We also found the gain of transfer learning increases as the training data decreases, and the convergence rate with transfer learning is better than that of the random initialization.

The field of artificial intelligence (AI) has created computers that can drive cars, synthesize chemical compounds, fold proteins and detect high-energy particles at a superhuman level. However, these AI algorithms cannot explain the thought processes behind their decisions. A computer that masters protein folding and also tells researchers more about the rules of biology is much more useful than a computer that folds proteins without explanation. Therefore, AI researchers like me are now turning our efforts toward developing AI algorithms that can explain themselves in a manner that humans can understand. If we can do this, I believe that AI will be able to uncover and teach people new facts about the world that have not yet been discovered, leading to new innovations.

![Figure][1] The McPherson 234/302 compact vacuum-ultraviolet spectrometer has a digital grating drive for precise wavelength selection and positioning from 30 nm to 1,100 nm. Micrometer adjustable slits vary from 0.01 mm ~3 mm in width and 2 mm ~20 mm in height. Software is available along with LabVIEW drivers. This instrument's normal incidence design has optional multiple input or output ports. It can also be easily used as a spectrograph with a microchannel plate intensifier or charge-coupled device detector, or as a scanning monochromator—one instrument can do both functions while remaining under vacuum. There are many options for customization: We can provide ultrahigh vacuum nonmagnetic versions or customized adapters for the customer's vacuum pumps, detectors, or light sources. Double monochromators for exceptionally low stray light and high spectral purity are also available. Special or standard, every instrument ships with certified spectral calibration. Compatible with all 2D-barcoded tube racks in SBS format and with a footprint of only slightly more than one plate/rack position, the Ziath DataPaq Express offers a space-saving way to integrate a fast full-rack scanner onto a liquid-handling robot. Designed with a separate power and processing box that can be positioned under your liquid-handling robot and under your deck, this compact scanner frees up vital deck space. Its uniquely low form factor allows liquid-handling robots easy gripper access to simply pick up and dispense from racks on top of the scanner. Offering rapid image scanning and decoding in just 2 s, the camera-based DataPaq Express will also help improve your robotic workflow. Baseplates and drivers are available to enable easy integration with most commercial liquid-handling robots. The Smart Evaporator C1 from BioChromato is an easy-to-use, affordable system optimized to concentrate or dry single samples directly from any tube or vial (up to 32-mm neck diameter) in even high-boiling solvents such as DMSO, DMF, or water. Drawing on BioChromato's patented spiral plug evaporation technology, the compact, benchtop system offers fast, effective evaporation in tubes or vials without solvent bumping, eliminating the risk of sample loss and cross-contamination and saving valuable time. The Smart Evaporator C1 can handle solvent volumes up to 40 mL, which can be extremely useful for concentrating compounds after organic synthesis or for drying analytical samples at relatively high speeds. The versatile C1 can also manage small tubes and vials (e.g., 1.5 mL) where solvent volumes can be as little as 0.1 mL or less. Porvair Sciences provides a complete design and manufacture service to help customers develop new and innovative custom microplates for specialist applications. We are widely recognized as a leader in the field of molding ultrapure plastic materials such as polystyrene, polypropylene, and polycarbonate. Decades of experience in ultrasonic welding, surface treatment techniques, co-sintering of polymers/silicas, and specialist assembly, combined with a strong understanding of analytical applications, make us an ideal OEM partner for development and production of optimized custom microplate solutions. From single-well to 1,536-well microplates, Porvair Sciences has the knowledge, expertise, and flexibility to design and manufacture to customer specifications. Our team of engineers and creative thinkers allows us to develop high-quality products for filtration, storage, and separation and to push the boundaries of microplate design for the life science and analytical markets. With its unique safety-locking mechanism and robust, adjustable support frame/lifting platform option, the Multicell PLUS high-pressure reactor from Asynt sets a new benchmark for operator safety, all-round accessibility, and ease-of-use. Manufactured from 316 stainless steel, the unit operates at pressures up to 50 barg and temperatures up to 200ºC. Asynt offers options for the system to be manufactured from alternative materials that can withstand highly corrosive/caustic chemicals, and for increased operational conditions up to 200 barg and temperatures of over 300ºC. While the Multicell PLUS accommodates 8 × 30 mL cells as standard, options are offered for 4-, 6-, and 10-cell arrangements with individual cell volumes up to 100 mL. Motor-driven, magnetically coupled overhead stirring is also offered as an option for more viscous reaction mixtures. Optional independent isolation of each cell allows the user to charge each vessel with differing chemistry and pressures without cross-contamination between cells. Milo is the world's first automated single-cell Western (scWestern) platform. The instrument measures protein expression in thousands of cells in a single run, allowing you to profile heterogeneity in your samples through single-cell analysis. Just load your cell suspension, and the scWest chip captures ~1,000 single cells. Milo then performs a fast, 1-min SDS-PAGE (sodium dodecyl sulphate–polyacrylamide gel electrophoresis) separation on each single-cell lysate on-chip. Then just probe with your favorite conventional Western blot antibodies to measure ~12 proteins per cell using a variety of multiplexing strategies. Milo's Single-Cell Western technology unlocks the single-cell proteome to measure more of the proteome than is possible with any other single-cell protein analysis technique. [1]: pending:yes

Machine learning has been widely adopted to accelerate the screening of materials. Most existing studies implicitly assume that the training data are generated through a deterministic, unbiased process, but this assumption might not hold for the simulation of some complex materials. In this work, we aim to screen amorphous polymer electrolytes which are promising candidates for the next generation lithium-ion battery technology but extremely expensive to simulate due to their structural complexity. We demonstrate that a multi-task graph neural network can learn from a large amount of noisy, biased data and a small number of unbiased data and reduce both random and systematic errors in predicting the transport properties of polymer electrolytes. This observation allows us to achieve accurate predictions on the properties of complex materials by learning to reduce errors in the training data, instead of running repetitive, expensive simulations which is conventionally used to reduce simulation errors. With this approach, we screen a space of 6247 polymer electrolytes, orders of magnitude larger than previous computational studies. We also find a good extrapolation performance to the top polymers from a larger space of 53362 polymers and 31 experimentally-realized polymers. The strategy employed in this work may be applicable to a broad class of material discovery problems that involve the simulation of complex, amorphous materials.

Computer-aided process synthesis has been an important field of chemical engineering for decades [2]. There exists a vast amount of methods in computer-aided process synthesis, in which the roles of human and computer are quite different and vary in their proportions. On one end of the spectrum, humans invent flowsheets, provide mechanistic models of apparatus and physicochemical properties, and employ computers solely in simulations to evaluate and check the invented designs. On the other end of the spectrum, there is automated flowsheet synthesis, which we call rather human-aided process synthesis by a computer. Therein, the structure of the process and operating levels are chosen autonomously by the computer based on input by the human (typically a problem statement and the physicochemical property data). Siirola [3] classified automated flowsheet synthesis into three categories: superstructure optimization, evolutionary modification and systematic generation. In superstructure optimization, a large flowsheet structure (the superstructure) is set up in a way, so that a large set of process alternatives can be obtained by removing parts of that structure [4,5]. An objective function or cost function is defined and the optimal configuration for the flowsheet is determined by an optimization algorithm that uses decision variables to remove parts of the superstructure. Evolutionary modification works as follows: A process flowsheet is devised (by any method at hand), analyzed and changed in one or more ways repeatedly to improve it.

Classification algorithms have recently found applications in computational physics for the selection of numerical methods or models adapted to the environment and the state of the physical system. For such classification tasks, labeled training data come from numerical simulations and generally correspond to physical fields discretized on a mesh. Three challenging difficulties arise: the lack of training data, their high dimensionality, and the non-applicability of common data augmentation techniques to physics data. This article introduces two algorithms to address these issues, one for dimensionality reduction via feature selection, and one for data augmentation. These algorithms are combined with a wide variety of classifiers for their evaluation. When combined with a stacking ensemble made of six multilayer perceptrons and a ridge logistic regression, they enable reaching an accuracy of 90% on our classification problem for nonlinear structural mechanics.