Minisymposium

Accelerating the discovery of new molecules and materials, as well as green and sustainable ways to synthesize and manufacture them, will profoundly impact the worldwide challenges in energy, sustainability, and healthcare. The current human-dependent paradigm of experimental research in chemical and materials sciences fails to identify technological solutions quickly. Recent advances in reaction miniaturization, automated experimentation, and data science provide an exciting opportunity to reshape the discovery and development of new molecules and materials related to energy transition and sustainability. In this talk, I will present a 'self-driving fluidic lab (SDFL)' for autonomous discovery and development of emerging advanced functional materials and molecules through integrating flow chemistry, online characterization, and machine learning (ML). I will discuss how modularization of different chemical synthesis and processing stages in tandem with constantly evolving ML modeling and decision-making under uncertainty can enable resource-efficient navigation through high dimensional experimental design spaces (>1020 possible experimental conditions). Example applications of the SDFL for the autonomous synthesis of quantum dots and specialty chemicals will be presented to illustrate the potential of autonomous robotic experimentation in reducing synthetic route discovery timeframe from >10 years to a few months.

For the past two decades, computational workflows and workflow management systems have enabled the automation of complex computing applications. They have empowered scientists to describe their computational problems in terms of tasks that need to be executed, the data they need to process, and in which order the tasks need to be performed. Because of the workflow formalism and the automation provided by workflow management systems, scientists in a number of domains, including bioinformatics, astronomy, earthquake science, and gravitational-wave physics have been able to achieve breakthroughs otherwise not possible.

Recently, experimental workflows in biology, chemistry, and material science are being formalized and their execution is being enabled via automation provided by cloud and self-driving labs. However, there is currently no connection between computational and experimental workflows.

This talk explores the synergies between computational and experimental workflow systems and how they can potentially complement each other to achieve scientific objectives.

The rapid advancements in cloud computing and the integration of experimental facilities, including self-driving labs, have resulted in an era where scientists can generate unprecedented amounts of data and conduct more extensive analyses across various scientific domains, including chemistry, materials sciences, molecular biology, and drug design. This capability enables a broader exploration of natural phenomena but also introduces significant challenges in effectively composing and scaling dataflow pipelines. This talk addresses these challenges by presenting innovative solutions for optimizing dataflow pipelines across cloud resources, thereby enhancing the study and application of scientific dataflows.

This talk will cover three main research components of our work when optimizing dataflow pipelines from self-driving labs to the cloud. First, we establish a taxonomy of common dataflow motifs ranging from simple producer-consumer pairs to complex multi-scale pipelines, applying these motifs to real-world use cases. Second, we discuss methods to mitigate data loss and pipeline inefficiencies, especially those arising from disparities in moving pipelines traditionally executed on high performance computing systems to the cloud. Last, we highlight our efforts to train and build a community of experts, emphasizing the development of tailored data analytics material across scientific domains.

Cloud environments have democratized access to large compute centers with easy to use interfaces and straightforward cost models. While this has lowered the barriers to many scientific disciplines to make reproducible computational experiments and data analysis, many of these enabling technologies can hurt reproducibility efforts inadvertently. This talk will explore some of the common interfaces, such as Jupyter notebooks and containers, to explore how these tools both support and hurt reproducibility efforts and offer ideas on some ways to mitigate these challenges.