Proceedings of a Workshop
Empowering Tomorrow’s Chemist
Laboratory Automation and Accelerated Synthesis
Proceedings of a Workshop—in Brief
INTRODUCTION: NEXT GENERATION OF CHEMISTS
The National Academies of Sciences, Engineering, and Medicine’s Chemical Sciences Roundtable convened a workshop on how best to automate the laboratory and train the next generation of chemists, chemical engineers, materials and formulation scientists, and others. This workshop explored the benefits of laboratory automation and opportunities to make this space more accessible for resource-limited institutions and identified ways to prepare the science, technology, engineering, and medicine (STEM) workforce. Panel sessions gathered together researchers from industry, academia, and government to focus on the various aspects of automation in the chemical sciences, including the challenges and benefits of automation in industry, automation training and development of curricula in academia, and the future of automation in the industrial laboratories and universities. Three potential paths forward emerged from the workshop discussions: creation of an online “living textbook” for laboratory automation resources and training; developing a “lab in a box” for students to learn the fundamentals of laboratory automation; and establishing a consortium of researchers to continue to share ideas on how best to harness automation to benefit industry, academia, and society.
POINT-COUNTERPOINT: WHAT WILL THE FUTURE OF CHEMICAL LABORATORY AUTOMATION MEAN FOR SOCIETY?
Martin Burke from the University of Illinois at Urbana-Champaign and Nicola Pohl from Indiana University Bloomington opened the workshop with a discussion about the promise and challenges presented by developments in laboratory automation and implications for the future. Burke shared a provocative question previously posed to him about the potential for automation to eliminate jobs. Both Pohl and Burke observed that, rather than outsourcing jobs, automation could amplify scientists’ work in highly significant ways. For instance, robotic assistance makes ambitious goals more accessible and expands scientists’ capacity to think creatively; furthermore, automated robotic systems create reliable, reproducible results that scientists can leverage for problem solving in novel ways. Burke and Pohl contemplated the benefits of centralized and decentralized automated processes. Centralized facilities may be filled with robots, but scientists could design automated instruments that are inexpensive and easy to deploy in smaller laboratories that lack access to centralized facilities with automated systems. Burke noted that peptide synthesis provides an example of both centralized and decentralized automation. First incorporated into the workflows of established companies, peptide synthesis is now performed by students in laboratories; instead of stirring a compound in a flask, students can program an amino acid sequence on a computer. Pohl envisions a shift from concentrating innovative work within a single laboratory to utilizing automated features in laboratories across the United States to develop new chemistry.
Burke invited participants to consider how automation enables researchers to think differently about synthesis. Chemistry could be generalized and modularized in entirely new ways; automated machinery could be combined
and customized not only to make molecules but also to carry out various processes, such as producing molecular functions. Pohl noted that automation can be used to develop new chemistries that are beyond the physical capabilities of humans, such as quickly altering states of matter to produce analyses that are not possible with current methods. Burke observed that automation also makes some chemistries safer and therefore researchers can rethink approaches to ensuring laboratory safety; for example, in flow chemistry, small molecules can be created quickly in safe, controlled environments.
Accessible, reproducible, interoperable chemistry that laboratories can use across the world present ethical questions. Although enthusiastic about the positive implications of automation, Burke and Pohl acknowledged the potential for harmful applications. Pohl noted that the 3D printing of ghost guns (firearms assembled from kits that do not carry serial numbers) and the production of illicit drugs provide examples of how automation could be misused. She explained that industries led international consortia to formulate policies and safeguards to prevent the misuse of DNA sequencing when it was first introduced. Given the speed of innovation, both panelists agreed that similar action could also be taken for automation, and sooner rather than later.
The impetus of this workshop, Burke observed, was the need to train a future workforce capable of harnessing automation’s potential. Untold numbers of people who may have groundbreaking, innovative ideas are not engaged in the chemical sciences because of barriers in the education system. For example, organic chemistry courses may tend to “weed out” students, but automation of chemical syntheses could lower barriers and bring additional promising minds into the field. Robotics could also lower barriers by reducing the amount of manual dexterity needed for laboratory work. Burke and Pohl pondered the best education level to introduce automation concepts into curricula to optimize student exposure, adding that computational data analysis is a particularly important skill. Burke observed that automation provides an opportunity to rethink the teaching of synthesis and chemistry and may cause a shift toward the teaching of chemical functions over structure. Pohl and Burke mentioned that additional workshops could be planned to bring together educators to reimagine chemistry education in this moment of rapid innovation.
SESSION 1: THE FUTURE OF LABORATORY AUTOMATION IN INDUSTRY
A panel of representatives from the chemical industry discussed their experiences with automation in the laboratory, how automation is shaping developments in the field, and the training and skills that future chemists might need. Spencer Dreher from Merck opened the session by characterizing automation at Merck as a holistic effort, which encompasses the complete DMTA (design, make, test, and analyze) cycle. He observed that automated chemistry, purification, analysis, and design systems provide rich data that enable researchers to improve tools as they use them and to enhance processes by applying machine learning. Matthew Bio from Snapdragon Chemistry, explained that his company uses automation to manufacture new chemicals to incorporate into clinical and commercial supply chains. Bio remarked that the experimental stages of drug development require that automation platforms be flexible, highlighting the importance of developing new methods of automation. Such flexibility is supported by new technologies, such as single board computers and user-oriented graphical automation platforms, which accelerate the development of automated systems early in the manufacturing process. Benji Maruyama from the Air Force Research Laboratory layers autonomy onto automation, enabling robots to make sequential decisions about what synthesis operations to conduct next, freeing researchers to focus on other tasks. He also observed that the use of advanced equipment should not be limited to the centralized laboratory setting; an unlimited number of talented individuals could spur new innovations using these tools in decentralized laboratories. To increase individuals’ access to chemical synthesis, Maruyama’s laboratory invented a low-cost, semi-autonomous, 3D printer that teaches itself how to make simple structures. All of the designs and operating systems are open source so that students and enthusiasts can try new algorithms. Ying Wang from AbbVie identified flow chemistry and high-pressure chemistry as examples of processes that leverage the benefits of automation for industrial processes. Wang commented that automation allows small groups of full-time chemists to produce large amounts of chemicals, which reduce costs and could increase access to diverse compounds for more researchers in the field.
Panel members offered specific examples of how their respective industries use automation. Dreher views automation as a holistic approach in pharmaceuticals because several different processes are automated, including the production of molecules, testing, and analytics. He highlighted automation as particularly crucial for collecting useful information for decision making. In Maruyama’s laboratory, his team programmed a robotic platform that could follow instructions to produce carbon nanotubes via carbon vapor deposition. This technique required the remote control of lasers, microscopes, spectrometers, and other laboratory equipment. Once the system was completed, the robot could autonomously analyze its products using diagnostic tools and read a database of past actions with a programmed goal. With this automated robotic system, Maruyama’s team could optimize decision-making processes and test hypotheses in the fewest possible number of experiments. Bio described an algorithm for process optimization used in manufacturing, where automated tools are used in safety assessments because they can systematically analyze for risk, reduce costs, and increase efficiency. Bio’s team employs off-the-shelf equipment and software to automate processes
as much as is practicable with existing laboratory tools, demonstrating that automated systems do not have to be purpose-built to provide value. Wang identified several ways that AbbVie uses automation in, for example, the custom building of reaction plates, processes for synthesizing compounds, and photo readout instruments.
The panel also considered research gaps and the potential challenges to addressing them. Several of the panelists recognized the difficulty of attracting talented coders to the chemistry field because of intense competition more broadly in the software field. Addressing specific research gaps, Dreher noted the need for soluble reagents that can be dosed in solution. He also expressed an interest in novel research to address the challenge of working with inorganics and metals that form heterogeneous mixtures. He added, however, that analytics is the most significant bottleneck, because chemists can currently run many more experiments than they can analyze; calculating a yield for a reaction without a product standard in less than 1 second would be a laudable goal. Maruyama agreed that this analytical bottleneck is a significant barrier in polymer chemistry and material sciences, as well. Other barriers include a lack of standardized software and hardware and the high costs of developing mechanical behavior. Wang highlighted diazomethane flow systems as an area of promising innovation, which could reduce hazards and increase the capacity for experimentation. She also expressed the need for inexpensive, reliable alternatives to automated bench experimentation tools, such as the H-Cube, an automated flow chemistry platform. Bio observed that universal sensors that can detect when bottles are full or empty could advance the chemical industry in appreciable ways. More fundamentally, the use of electronic equipment in flammable, solvent environments is a ripe area for innovation.
The panelists also described the limits of current automated technologies, including those that they seek to work beyond and those where chemistry is simply resistant to automated processes. Maruyama noted that in materials science, changes in processes employed by humans happen with relative ease, but changes to automated processes are more resource intensive. Compartmentalizing automated processes to make them interchangeable provides more flexibility in experimental design. Dreher remarked that automation is useful for performing simple processes repeatedly, but not for performing new ones. He believes advances in robotics are needed to make automation more agile and responsive to novel approaches to experimentation. Wang commented that in the pharmaceutical industry, automation shapes what reactions are employed. Generally, Wang noted, chemists rely on automation to conduct automated and accelerated experiments in prevalent chemistry, rather than using them to innovate. For example, AbbVie uses automation for the most common types of reactions, including cross couplings and other amenable parallel transformations, which produce 20 to 25 percent of the company’s registered compounds. She observed that the use of automation for common chemistry improves throughput and is highly efficient, but creates the risk of bias toward the automated methods in current use.
Wang added that automation in instrumentation offers significant advantages, but further innovation is needed to address current limitations. Bio proposed that research could be conducted with sets of molecules that can be produced within the current limitations of automated systems in order to focus on innovation. Dreher suggested that more expansively exploring all of chemical space can help standardize molecule production in ways that will advance the field of precision medicine; for example, where many similar molecules must interact in a myriad of ways. However, some reactions, such as debenzylation and parallel nitrate reduction, cannot be automated safely or easily, according to Wang. Maruyama observed that any process involving image analysis is difficult to automate and integrate into workflows. With spectral analysis, for example, a person must verify results. Maruyama asserted that automation may not be directed toward removing humans entirely, but instead to producing more efficient teams of robots and researchers that can work synergistically.
The panel contemplated ways to democratize automation in industry to provide researchers with greater access to emerging technologies to stimulate innovation. Maruyama contended that decentralization of automation and greater access to maker spaces where people could tinker with chemistry would increase society’s research capacity. Niche research areas, such as orphan drugs, benefit from lower barriers to research. Bio believes that the costs of entry to research and development need not be prohibitive; his company started with small numbers of single-board, networked computers, which were adapted for complex automated laboratory work. Dreher noted that Merck is innovating automated tools for centralized use and disseminating them for decentralized use. Maruyama and Dreher affirmed that precompetitive, standardized approaches to supporting tools across industries would foster collaboration and innovation among researchers. With so many tools and platforms, none are well supported and simplifying their use for scientists outside of particular niches is difficult. Dreher remarked that academics with innovative ideas often face challenges obtaining funding from industries. Wang noted that academics attempting to prove technology is worth the investment to industry could provide highly persuasive data. Maruyama added that academics seeking adoption of an innovative technology by industry could demonstrate the low risk of investment in that technology.
The panel recognized that automation creates a need to update training and curricula. The panelists discussed their experiences searching for job candidates, which might inform upstream training for students. Bio noted that people who have experimented with the functionality of various types of hardware, as well as those who have coding experience, are attractive candidates; however, his team also seeks people who are open to, even if lacking
training in, automation in chemistry. Candidates with experience in cross-functional collaborations across the fields of chemical engineering, analytical chemistry, and software development are particularly valuable, and Bio suggested that interdisciplinary teams in academia could thus provide effective training environments. Dreher agreed and added that the ability to communicate across disciplines, such as between chemists and software engineers, is a sought-after skillset. Maruyama noted that candidates should have a general understanding of the significant automation algorithms in current use and of their functions. The panelists also discussed additions to curricula that could expose students to automation. Wang suggested that students set up parallel reactions so that they can experience parallel means of conducting experiments and optimizing their approaches. She added that these lessons require minimum investment in instrumentation for school laboratories. Wang also suggested that students work with a simple flow setup to understand the basics of running reactions using flow technology. Bio concurred and remarked that continuous manufacturing and continuous flow typically require automation, which makes these processes attractive topics for curriculum development.
The panel concluded with some overall observations about automation. Wang asserted that automation could fundamentally alter how chemistry is performed in the laboratory. The day when every bench chemist utilizes a suite of laboratory automation during daily synthesis is when automation will be integral to chemistry, she added, which will require commitment and planning. Dreher emphasized the ongoing importance of human–machine synergy, which will push industry to modernize and optimize automated systems to improve their businesses. He predicted that many more synthetic chemists will work in the field in the future. With the potential for more newcomers to the field, Bio is enthusiastic about the prospect of democratization and pondered how best to engage new innovators to solve problems. Regardless of the ways in which the future workforce engages with industry, Maruyama observed that a great deal of chemistry remains hands on, and more utilization of post-industrial tools is needed to advance the field. He looked forward to a future when an artificial intelligence (AI) scientist wins the Nobel Prize in chemistry, which provides a motivating goal for present and future scholars.
SESSION 2: THE FUTURE OF LABORATORY AUTOMATION IN ACADEMIA
A panel of representatives from academia gathered for the second session to discuss their experiences with automation and their ideas about teaching automation to students to prepare them for the chemistry laboratory workforce. Klavs Jensen from the Massachusetts Institute of Technology described his work in accelerated molecular development through the integration of molecular prediction, generative models, and computer-aided synthesis planning. Jensen highlighted a recent automation project, which involved a 96-well reaction plate in a closed loop system assisted by AI that can define molecular targets, plan the necessary chemical synthesis, and then synthesize the characterized compounds using robots. Echoing comments from the first session, Jensen noted the challenge of working with various types of analytical software, few of which have been allowed to mature in a crowded marketplace. Lee Cronin from the University of Glasgow focuses on standardizing chemistry to increase precompetitive access to synthesis through chemputation, the universal code–enabled control of chemical reactions using a standard programming language and ontology. His team built a chemical state machine that allows for interoperability, uses a universal code, and removes ambiguity in the comparison of experimental results from different laboratories by clarifying their contexts of development. Cronin stressed the need for a more reliable set of tools in the automated chemistry space. Andy Cooper from the Materials Innovation Factory and the Leverhulme Research Centre for Functional Materials Design at the University of Liverpool is developing the autonomous mobile robotic chemist. This approach automates the researcher, by programming a robot to autonomously perform experiments on mostly unmodified laboratory equipment.
Nicholas Ball from Pomona College focuses on curriculum development for undergraduate and graduate students and the stage at which students should be exposed to automation concepts. Next, Mimi Hii from Imperial College London, who directs its Centre for Rapid Online Analysis of Reactions (ROAR), outlined how ROAR helps ease the analysis bottleneck in current automated approaches by training students to process large amounts of data and accelerate research in automated methods. ROAR utilizes robots, flow chemistry platforms, and high throughput experimentation reaction sets, and sponsors a program to train post-graduate students in next-generation synthesis and reaction technology.
Rachel Switzky from the University of Illinois at Urbana-Champaign works with data scientists, AI engineers, and chemists in multidisciplinary, collaborative projects using a process of design thinking, a method of locating and engaging relevant experts in the development of various applications. Switzky promoted the idea of democratizing molecule-making beyond specialized applications, which should be user-centered to lower barriers to scholars and researchers. Jin Cha from the National Science Foundation (NSF) joined the panel to learn more about how NSF can support automation in chemical synthesis.
The panel considered the question posed by Burke and Pohl: “Will automation take my job?” Cronin responded that, rather than eliminate jobs, the standardized and automated production of molecules will redefine the work of chemists, shifting their focus to design and other creative endeavors. He observed that by changing the nature of
chemical synthesis, automation will expand capabilities well beyond their current limits—potentially leading to the exponential growth of molecules, similarly to how Moore’s Law predicted the expansion of transistors. Instead of taking jobs, Cronin believes, automation will likely create more jobs globally. Switzky agreed that automation in synthesis will lead to job growth and allow more people to engage with the field. She perceives educators’ primary goal as removing barriers to make this career pipeline easier to navigate. Hii remarked that the nature of necessary skills and training will shift with changes in the field. She uses the acronym AUTO (analysis, understanding, technology, optimization) to define the major skills and training needed for chemical synthesis. Hii uses AUTO principles in her program in London, which industries in both the United Kingdom and the United States believe will be applied by future researchers. Examples of each of the four areas include the use of analytical tools such as spectrometry and chromatography (analysis), identification of reactive intermediates and advanced synthesis (understanding), kinetics and thermodynamics and reactor designs (technology), and experimental design and parallel experimentation (optimization).
As with their industry counterparts, the panelists from academia discussed the merits of automation in decentralized and centralized contexts. Cooper extolled the opportunities for distributed experiments in groups of decentralized laboratories working in concert, noting that his laboratory in Liverpool provides remote access to international researchers much like Hii’s laboratory in London. Jensen supported the virtues of decentralization, remarking that academic laboratories can build their own automated systems and then collaborate with other groups of students worldwide. He noted that the costs of automated systems do not have to be prohibitive; the costs of 3D printers have been declining, and the systems that enable these and other automated equipment to communicate with their counterparts in other laboratories can be inexpensive. The challenge to fully realizing decentralized automated chemistry, Cronin reiterated, is interoperability: chemists not only need data in standardized formats for effective collaboration but also must understand the context in which those data are generated. He added that he is working with an international group of researchers to examine how code for generating molecules works in automated systems used in academic institutions in Canada, the United States, and Scotland. Hii observed that centralized facilities can invest in state-of-the-art equipment, which can be used by stakeholders from other institutions who lack the capital to build their own systems. These centralized systems can democratize access to automated systems, but only if people use them. In contrast, Hii noted, decentralized automation can happen in laboratories on college campuses all over the world. For example, during the pandemic, students in an undergraduate chemistry class in the United Kingdom received an at-home kit to build a simple UV-Vis spectrometer out of LEGO blocks. The goal of the lesson was for students to learn to use the Python programming language to create feedback loops, which can help prepare them for environments with more complex programming languages and machinery later in their careers.
Regarding the goal to improve student understanding of automation, the panelists contemplated potential additions to their undergraduate programs. Ball acknowledged that many departments that are not well funded tend to be conservative about faculty hiring, which complicates efforts to attract talent and build the pipeline in data science, computational knowledge, and automated applications. He remarked that undergraduate education is broadly organized as a buffet of courses, with limited administrator interest in introducing automation into curricula; however, helping students build fundamental knowledge in instrumentation, including setting up systems and data analysis, could have a high impact. Switzky is building an ecosystem of tools, including a digital molecule maker, for use by learners at the K–12 and undergraduate levels; the molecule maker is being piloted at the University of Illinois at Urbana-Champaign for implementation in undergraduate courses. Switzky contended that tools such as the digital molecule maker can help shift the paradigm from one where traditional organic chemistry poses a barrier to students, to another where universities can build a pipeline of potential molecule designers.
Both undergraduate and graduate programs could do more to advance interdisciplinary training, which the panelists agreed is a fundamental aspect of automation. Cronin remarked that instructors could teach chemists the principles of chemical engineering; likewise, they could teach engineers synthetic chemistry. Hii connected interdisciplinary training to team-based education. She noted that high-throughput robotic data enable each student in a classroom to work with a different set of conditions, which could be combined to teach students about statistical analysis. She underscored that the analysis of datasets is the future of chemistry and that the analysis of multiple reactions under multiple parameters is an industrial norm that students could be able to experience. Furthermore, collaborative teamwork models can yield better results in the classroom, compared to individualized, competitive course design that rewards students who are primarily good at laboratory work. Ball added his support for collaborative learning models, which are more explicitly inclusive and are becoming the expectation for scholars in future pipelines. Ball described a model in which an instructor shares a research question and data with students and invites them to problem solve together, which is how chemistry is typically done in professional settings. Ball and Cooper suggested that de-siloing academic disciplines and fostering collaborative problem solving are needed to foster collaboration at both the undergraduate and graduate levels. Jensen suggested that graduate programs be deliberate about creating multidisciplinary work groups.
A multidisciplinary approach in automation emerged as a key concept during the panel discussion. Switzky emphasized the need to understand the reluctance of some scientists to collaborate and the barriers that prevent them
from doing so. She noted that one obvious barrier is that chemistry has its own language, which could be made more accessible and focused on the needs of learners. This point applies to the education of new chemists and their peers; Switzky echoed earlier comments that data scientists need help understanding the language of chemistry. Cha noted that certain NSF funding mechanisms support both single and multiple investigators from different disciplines and that some awards in the NSF chemical synthesis program have automation as a central component. The Centers for Chemical Innovation also award multi-institution grants for innovative research. Cronin suggested the creation of workshops where chemists learn to code and data scientists learn about chemical reactions. The goal, he emphasized, is to encourage diverse groups of students to think in fundamental ways about both chemistry and coding, and to appreciate the challenges inherent in understanding and developing automated systems. Jensen reflected that students need a learning environment in which people from different disciplines build trust and work together on problems. Cooper observed that one barrier to these collaborative models is the results-focused mindset that is embedded in some parts of scientific culture. The focus for the students, he added, would simply be learning to do new things.
The panelists turned to the largest challenges that they have encountered when implementing automation in their respective departments. Jensen reported that the lack of a shared language among fields remains an issue, although he has successfully collaborated with several organic chemists. Ball noted that the competitive pressure at R1 institutions1 may be stifling innovation. Outside of R1 institutions, instructors teach in environments in which the pressure to write grants to fund their work is lower. This career path is fundamentally different, Ball observed, with more freedom and opportunities to engage in innovative, multidisciplinary endeavors. Cooper suggested that a key challenge is hiring a blend of faculty with the different skillsets and backgrounds needed to teach the multidisciplinary nature of automation. Hii added that this challenge is particularly relevant in the context of data analysis. Although the adoption of automation per se does not present significant barriers, it does raise questions about how to analyze the growing volumes of data. This endeavor could benefit from collaboration, not only with data scientists and machine learning experts but also with students who have been exposed to automation concepts.
Cronin observed that, over the next decade, academia may be forced to eliminate some of the fundamentals currently taught in schools to adjust to innovations in automation; furthermore, some cultural traditions in laboratory work may be replaced with the use of more advanced analytical tools as they become available. Ball also cited entrenched ways of thinking in academia as a significant challenge. He recounted a controversy regarding the publication of an article in Angewandte Chemie2 about how diversity efforts were harming science; to the contrary, he noted that diverse modes of thinking are central to fostering innovative mindsets at the undergraduate level. He suggested a phenomenological approach to instruction, in which an instructor discusses a reaction without explaining how it works to see what questions students pose. Overall, Ball emphasized that educators could consider the human element, and if chemistry has a reputation for being intimidating, then greater effort is needed to welcome students and colleagues.
The panelists concluded their session with a brief discussion of ethical considerations in chemistry instruction. Switzky affirmed that educators have a responsibility to include ethics in instruction. In educational contexts, students could be engaged in conversations about unintended consequences and unanticipated events. She also noted that instructors must be mindful of their own biases in order to lead effective conversations about ethical considerations in science. Cronin observed that tragic consequences can happen if software design does not provide safeguards within automated systems—such as the ability to override an automated cockpit door in an emergency. Verification of how users employ tools for benevolent (or less benevolent) purposes is equally important. He added that encryption and other measures can help provide a layer of security. Hii recognized that researchers cannot anticipate all of the uses of chemistry for nefarious purposes; however, they can train people who want to act ethically to avoid pitfalls. Her own program requires training in responsible research and innovation, which provides an opportunity for students to consider the possible negative effects of their work. Cooper provided perspective, remarking that automation is not the largest ethical question in chemistry compared to the equitable supply of medicines and the production of greenhouse gasses. As automated tools become more prevalent, he observed, ethics become more important.
During day 1 of the workshop, 15 groups of researchers presented posters3 on various topics regarding automation in the chemical sciences. Participating universities and institutions included (but were not limited to) the Ames Laboratory, Carnegie Mellon University (CMU), the Czech Academy of Sciences, the Defense Advanced Research Projects Agency, Haverford College, the H.E.L Group, Indiana University, Michigan Technological University, the National Insti-
1 R1 institutions are defined by the Carnegie Classification of Institutions of Higher Education, see https://carnegieclassifications.iu.edu/classification_descriptions/basic.php.
2 This article was withdrawn from Angewandte Chemie, see https://onlinelibrary.wiley.com/doi/10.1002/anie.202006717.
3 Poster presentations can be found online at https://www.nationalacademies.org/event/05-25-2021/laboratory-automation-and-accelerated-synthesis-empowering-tomorrows-chemist-a-workshop.
tute of Standards and Technology, Northwestern University, Texas A&M University, the University of Maryland, College Park, and Whitman College.
The workshop included three breakout sessions, each assigned a different goal related to modules, resources, and blueprints. The report-outs from the breakout sessions are summarized below.
Breakout Session 1: Out of the Box: Providing the Tools and Resources Needed to Prepare Undergraduates for Automation
Arsalan Mirjafari from Florida Gulf Coast University introduced the first breakout group. The group’s stated goal was to generate ideas about how to democratize synthesis with a low-cost, viable automation technology for use in education at the undergraduate level. To achieve this goal, this breakout group envisioned a “lab in a box,” or a contained group of lessons to expose undergraduates to the concepts of automation in the laboratory. The breakout group divided this task into two parts: “Ideation,” which concerns methods of introducing automation concepts in the classroom, and “Brainstorm,” which concerns the physical materials and prototypes that would be included in the laboratory in a box. Mirjafari reported that the ideation group explored the need to introduce automation in undergraduate curricula. Some faculty in this group believe that students will best learn to appreciate the difficulty of chemical synthesis through hands-on, traditional approaches. However, the group suggested that if automation were taught at the undergraduate level, then reproducibility would be a central concept. The ideation group considered operations that are both inexpensive and interesting to execute, such as automating a process using Raspberry Pi—a small, low-cost, single-board computer. One way to make lessons easy to execute is to leverage local resources available to schools wherever possible. Rather than revising whole courses, which is burdensome for educators, common laboratory assignments could be automated; however, the instructors would need to be comfortable with data and AI components.
The group also considered the value of introducing automation to younger students, including high school and middle school students, through age-appropriate projects. Mirjafari also reported on the deliberations of the brainstorm group, which considered cost and safety to be significant factors. This group identified several easy and safe projects for students, such as an auto-titration with orange juice; experiments that could be done in water to remove the need for organic solvents; and employing amino acid–derived materials for CO2 capture. The group noted that students could use nuclear magnetic resonance spectroscopy to analyze carbon peaks and experiment with carbon capture, which could be incorporated into an engaging lesson on climate change. The group discussed adapting the lab-in-a-box lesson for introductory chemistry classes because they have the potential to reach significantly more undergraduate students than upper-level courses.
An audience member suggested a project in which students pool data, which could be done on large scales and involve many schools, as an introduction to reproducibility and statistics. Mirjafari ended the presentation of the first breakout group with a call for volunteers to work on this project.
Breakout Session 2: Designing a Textbook to Implement Automation in Graduate Research and Funding Opportunities
Anne LaPointe from Cornell University introduced the second breakout group, whose charge was to plan the production of an open-source textbook for instruction in automated and accelerated chemistry. The group envisioned an online, living document that would be more accessible than a traditional textbook typically purchased by students for coursework. Content could be crowdsourced from a wide, multidisciplinary group of experts with chemistry, computer science, and engineering backgrounds. To appeal to broad audiences online, the textbook could be organized into modules based on skill level, from student to professional, with three main sections: (1) concepts, definitions, and methods; (2) unit operations, which includes hardware, software, and analytics; and (3) resources, including case studies, videos, and information on funding sources. The textbook could also include an annotated bibliography.
LaPointe noted that the first section could offer a general overview without favoring any particular discipline within chemistry. It could also provide a historical perspective on the development of automation to provide context for students. To supplement the general automation concepts and themes, the group suggested inclusion of ethical issues for readers to contemplate, as well integration of the concepts of diversity and inclusion into the first section. LaPointe framed the second section as a “choose your own adventure” for audiences with diverse skillsets and experiences to identify and experiment with relevant applications. The section could be divided into four categories: adding and removing chemicals; safety and ergonomics; analysis; and hardware, software, programming, and interfacing. LaPointe noted that crowdsourcing short videos on various applications could be useful, with the proposed content curated by experts and updated regularly to ensure quality. The third section of the living textbook could present case studies and appendixes. Case studies could address topics such as pharmaceutical optimization, academic studies with
creative workarounds, materials or polymers, machine learning, and flow chemistry. The group also considered potential resources, such as Journal of Chemical Education papers, short videos on methods, computing resources, open-source classes, and collaboration opportunities. LaPointe ended the presentation of the second breakout group with a call for volunteers to work on this project.
Breakout Session 3: Advancing Automation and Beyond: Creating a Blueprint for Consortia to Share Data and Learn from Each Other
Shane W. Krska from Merck Research Laboratories introduced the third breakout group. To sustain momentum on collaborative projects in chemical automation, the group endeavored to design a consortium of industry, academic, and funding partners that could aim to democratize chemistry by advancing the use of automation in undergraduate and graduate education. The group formulated a potential three-part action plan to assess problems, identify resources and roadblocks, and plan first steps. This plan aims to address the lack of widespread undergraduate and graduate training in automation and the potential for increasing the inequities of the digital divide as automation becomes more prevalent. Central to this plan are several key principles: exposure to the value and concepts of automation at an early stage could be a primary focus; automation provides an opportunity to increase inclusivity in the field; intentionality can be utilized to shrink the digital divide; and cross-disciplinary training and automation go hand in hand. This endeavor could draw on the expertise of skilled automation practitioners from industry, academia, and government laboratories, including cross-disciplinary teams of engineers, chemists, software developers, and project managers. Krska also identified funding streams for increasing connectivity, broadening the reach of science education, and developing the workforce.
The groups also discussed several roadblocks. For instance, intellectual property and contractual hindrances may inhibit collaboration; differing perceptions of automation also complicate efforts to formulate a shared language and vision; and conventional synthesis education, which focuses on traditional non-automated approaches, is often entrenched. To realize the vision for a consortium, the group suggested several next steps, which encompass searching for a highly visible sponsor to lend credibility and build support; chartering a possible cross-disciplinary roundtable under a sponsor; delivering on the products of the first two breakout groups (lab in a box and/or open-source textbook); and identifying new partners. Krska noted that questions remain about the project’s scope and industry partner involvement.
SESSION 3: THE PROMISE OF AUTOMATION
In the third and final session, a panel of individuals from a variety of fields presented projects that represent the potential of automation. Alán Aspuru-Guzik from the University of Toronto, who is the Canada 150 Research Chair in Theoretical and Quantum Chemistry, opened the session with a presentation on self-driving laboratories, or materials acceleration platforms (MAPs), which adopt an inverse design model that continuously uses data in real time to efficiently shape the functioning of an automated system. These automated systems rely on multidisciplinary approaches that require fundamental knowledge of molecular properties, photophysics, and the application of AI to filter superfluous reactions. The resulting automated workflows enable rapid screening; through the use of calculations based on density functional theory; for example, an approach that usually takes approximately 13 hours can be finished in approximately 15 minutes. He described one platform called Molecular Accelerated Discovery of Novelty-Enabled Systems, which is being used to identify a new molecular class: solid-state organic lasers. The system uses a multidisciplinary, two-step process to identify lead candidate molecules, including virtual screening based on quantum chemistry and experimental online validation. Aspuru-Guzik cited The Machine in Illinois, the Acceleration Consortium in Toronto, Chemputer in Glasgow, and N9 in Vancouver as experimental platforms with similar developing technological approaches.
Aspuru-Guzik noted that Bayesian optimizers are critical orchestration software used to bring together all of the parts of MAP systems, which incorporate the human–robot interface, AI, automated computations, online analysis, data management, and automated synthesis. Several Bayesian optimizing tools are currently in use, including Phoenics (Bayesian neural network optimization), Chimera (pareto multi-objective optimization), and Gryffin (categorical variable optimization). By employing these three tools, Aspuru-Guzik and other researchers can perform practical optimization. Newer tools have become available, such as Golem (robustness-aware optimization), Gemini (multi-fidelity optimization), and Olympus (benchmarking), which further increase the capacity for optimization. One use case for this technology involves optimizing power conversion efficiency of non-fullerene acceptors for organic solar cells. Aspuru-Guzik also provided an example of Bayesian tools that his team uses to optimize reactions produced by the self-driving laboratory. In response to audience questions, Aspuru-Guzik suggested that although molecules have been the initial focus of MAPs, they are ahead of polymers by only about 3 years. His team is already working to acquire a polymer synthesizer for further testing. He also noted that other systems like metal–organic frameworks can be produced and measured. Overall, he explained that optimized systems can screen for ease of synthesis, which enables robots to identify materials that are not difficult to produce.
Joseph DeSimone from Stanford University presented on the digital manufacturing technology used by his company, Carbon. Referring to earlier comments, he acknowledged the challenge of assembling people from different disciplines to work together, which is essential to developing automated systems. Carbon has used the principles of mechatronics engineering—an interdisciplinary branch of engineering focused on the integration of mechanical, electronic, and electrical engineering—to develop rapid 3D printing hardware. The printer uses software-controlled chemical reactions to apply oxygen and light to grow objects; therefore, this technology can be constantly improved via software upgrades. This technology can be applied to automotive, medical, and consumer products; the potential exists for super high-resolution printing, such as for the manufacture of transdermal patches for delivering vaccines.
DeSimone described several ways that digital manufacturing technology can be optimized. Generally, the more the technology is used, the more it is optimized. More specifically, a company’s business model can promote technology optimization: Carbon, for example, employs a subscription system whereby customers partner with the company and have a stake in improving the technology as they use it. Furthermore, with digitized printing processes, users can track each printed item from fabrication to performance, which generates analytic data to inform optimization efforts. DeSimone observed that digital fabrication provides scalability, supply chain agility, adaptability, and lower energy use than other manufacturing methods.
Rebecca Doerge from CMU presented on its Cloud Lab, which is an automated, remote-controlled laboratory capable of handling all aspects of daily laboratory work, from experimental design, data acquisition, and analysis, to sample production. Expected to be fully operational in fall 2022, the laboratory aims to help democratize science by offering researchers the ability to remotely access a fully automated laboratory that uses state-of-the-art technology and machine learning to iteratively optimize processes. In order to demonstrate the viability of the Cloud Lab, CMU sent a doctoral student to conduct research at the Emerald Lab, whose facilities are similar to those of the proposed CMU laboratory. Doerge explained that, using a traditional laboratory on the CMU campus, the student was able to synthesize three compounds in 1 week. Using the Emerald Lab, the student was able to synthesize hundreds of compounds within the same timeframe. Doerge described another proof-of-concept experiment involving automation and AI for polymer synthesis. This experiment was predicated on the belief that traditional MRI imaging would require researchers to test 50,000 monomer compositions over many years to locate an ideal imaging agent. Using AI systems proposed for the Cloud Lab, scientists found ideal candidates after testing fewer than 400 polymers in 1 week.
In the Cloud Lab, researchers can access automated instruments for materials handling, sample preparation, property measurements, cell preparation, organic synthesis, separations, spectroscopy, mass spectrometry, bioassays, microscopy, crystallography, and polymers/materials. The Cloud Lab is scalable, that is, more instruments can be added and workflows can be customized to accommodate any automated instrument. Doerge noted that the Cloud Lab offers the practical advantage of cost sharing with other institutions. Faculty do not need to travel and set up new physical laboratories, but can collaborate from wherever they are. She also noted that the Cloud Lab could benefit under-resourced institutions without access to advanced instruments. Responding to questions from the audience, Doerge noted that 50–70 research threads can be processed simultaneously, depending on the instrumentation used and the number of operators involved. The Cloud Lab can also produce materials in milligrams, grams, and kilogram scale.
Burke and Pohl closed the workshop by summarizing the lessons drawn from the workshop panels and discussions. Burke highlighted the participants’ readiness to turn ideas into action, by continuing to collaborate on the lab-in-a-box, textbook, and consortium concepts to democratize automation. By expanding access to innovative tools to a vast number of new researchers, automation could revolutionize the field. Pohl remarked that the conversation at the beginning of the workshop involved wondering whether automation would eliminate jobs; the irony is that nearly every conversation revolved around the benefits and challenges of bringing people together. Burke added that the scale of potential collaborative projects is global. Given the enthusiasm of participants and new initiatives emerging from the workshop, Pohl and Burke expressed optimism about future developments.
PLANNING COMMITTEE: Martin Burke, May and Ving Lee Professor of Chemical Innovation, University of Illinois at Urbana-Champaign; Tim Jamison, Professor of Chemistry, Massachusetts Institute of Technology; Shane W. Krska, Distinguished Scientist, Merck Research Laboratories; Anne LaPointe, Director, Catalyst Discovery and Development Laboratory, Cornell University; Robert E. Maleczka, Jr., Professor of Chemistry, Michigan State University; Arsalan Mirjafari, Associate Professor, Florida Gulf Coast University; Nicola L. B. Pohl, Associate Dean, Natural and Mathematical Sciences and Research, College of Arts and Sciences, and Professor of Chemistry and Joan and Marvin Carmack Chair of Bioorganic Chemistry, Indiana University Bloomington; Becky Matz, Research Scientist, Center for Academic Innovation, University of Michigan. National Academies’ staff were Linda Nhon, Jessica Wolfman, and Ayanna Lynch.
DISCLAIMER: This Proceedings of a Workshop—in Brief was prepared by Jay Weixelbaum, Science Writer at Rose Li & Associates, as a factual summary of what occurred at the workshop. The statements recorded here are those of the individual workshop participants and do not necessarily represent the views of all participants, the planning committee, the Chemical Sciences Roundtable, or the National Academies.
REVIEWERS: To ensure that this Proceedings of a Workshop—in Brief meets institutional standards for quality and objectivity, it was reviewed in draft form by Anne LaPointe, Cornell University, and Arsalan Mirjafari, Florida Gulf Coast University. We also thank staff member Tom Arrison for reading and providing helpful comments on this manuscript. The review comments and draft manuscript remain confidential to protect the integrity of the process.
ABOUT THE CHEMICAL SCIENCES ROUNDTABLE
The Chemical Sciences Roundtable provides a neutral forum to advance the understanding of issues in the chemical sciences and technologies that affect government, industry, academic, national laboratory, and nonprofit sectors and the interactions among them and to furnish a vehicle for education, exchange of information, and discussion of issues and trends that affect the chemical sciences. The roundtable accomplishes its objectives by holding annual meetings of its members and by organizing webinars and workshops on relevant and important topics.
Chemical Sciences Roundtable members are Linda Broadbelt (Co-Chair), Northwestern University; Michael J. Fuller (Co-Chair), Chevron Energy Technology Company; Brian Baynes, MODO Global Technologies; Michael R. Berman, Air Force Office of Scientific Research; David Berkowitz, National Science Foundation; Martin Burke, University of Illinois at Urbana-Champaign; Miles Fabian, National Institutes of General Medical Sciences; Laura Gagliardi, The University of Chicago; Bruce Garrett, U.S. Department of Energy; Franz Geiger, Northwestern University; Carlos Gonzalez, National Institute of Standards and Technology; Malika Jeffries-El, Boston University; Jack Kaye, National Aeronautics and Space Administration; Mark E. Jones, Dow Chemical (Retired); Mary Kirchhoff, American Chemical Society; Robert E. Maleczka, Jr., Michigan State University; David Myers, GCP Applied Technologies; Timothy Patten, National Science Foundation; Nicola Pohl, Indiana University; Ashutosh Rao, U.S. Food and Drug Administration; Sunita Satyapal, U.S. Department of Energy; and Jake Yeston, American Association for the Advancement of Science.
SPONSORS: This activity was supported by the National Science Foundation under Grant CHE-1546732 and the U.S. Department of Energy under Grant DE-FG02-07ER15872. Any opinions, findings, conclusions, or recommendations expressed in this publication do not necessarily reflect the views of any organization or agency that provided support for the project.
Suggested citation: National Academies of Sciences, Engineering, and Medicine. 2022. Empowering Tomorrow’s Chemist: Laboratory Automation and Accelerated Synthesis: Proceedings of a Workshop—in Brief. Washington, DC: The National Academies Press. https://doi.org/10.17226/26497.
Division on Earth and Life Studies
Copyright 2022 by the National Academy of Sciences. All rights reserved.