Chemistry Meets Code: Revolutionising Reactions with Computational Power

Artificial Intelligence (AI) has become a pivotal technology in 2024, influencing various sectors worldwide. Its rapid integration into industries such as finance, healthcare, education, and transportation are driving innovation and efficiency. Within chemical research and development, AI has made significant changes in areas like novel molecule design and predictive analysis, such as retrosynthesis. However, currently it still remains in its developmental stage, with much room for growth and refinement. There are significant challenges to overcome, such as developing AI into a more generalised tool capable of handling diverse and modern chemical reactions, as well as addressing issues related to unseen or new data. The societal impacts of AI also demands careful consideration, particularly its challenges surrounding data privacy, ethical implications, and potential biases.

In this quarter's newsletter, we explore the research taking place across the Midwest, spotlighting key individuals at the forefront of integrating AI into their scientific work. These researchers are not only trying to solve complex issues in their field but also navigating the challenges that come with implementing AI in chemistry. Their efforts highlight the critical role of chemical expertise in advancing AI-driven innovations. Let’s take a closer look at their work and the contributions they are making for future scientists.

So, where did it all begin? - AI’s influence in the field of chemistry traces back to the development of efficient computer technology in the 1940s, when the first solutions to complex wave equations for atomic systems became possible. By the early 1950s, the advent of semi-empirical atomic orbital calculations marked a pivotal moment, allowing theoretical chemists to harness the power of early digital computers. These advances enabled researchers to model and predict chemical behaviour with unprecedented precision, establishing computational chemistry as a reliable tool. Since their early adoption, computer models have become essential across various branches of chemistry, helping solve complex chemical problems. They are now crucial for tasks such as multiscale biomolecular simulations, which aid in identifying drug binding sites on target macromolecules, and virtual screening techniques like molecular docking, pharmacophore modelling, and quantitative structure-activity relationship (QSAR) studies. These tools have revolutionised drug discovery, enabled de novo drug design and optimise chemical reactions. In addition, the vast amounts of chemical information from open-access and commercial databases have been further developed and expanded through computational methods. Moreover, the integration of AI and data analytics has accelerated the discovery of eco-friendly materials, optimised plant design and management, and improved processes for sustainability.

It is important to highlight as well that the correct integration of AI in chemistry presents several challenges that researchers must navigate. One significant hurdle is the quality and availability of data; AI models rely heavily on extensive and high-quality datasets to make accurate predictions. In chemistry, gathering such data can be difficult due to the variability in experimental conditions and the complexity of chemical systems. Additionally, many AI algorithms function as "black boxes," making it challenging for chemists to interpret their outputs and understand the underlying mechanisms driving predictions. There is also the concern of reproducibility, as AI-driven approaches may yield different results under slightly varied conditions. Nonetheless, these challenges are well recognised and are currently being addressed through various approaches.

Figure 1. “A brief history of AI.” By Ryan Byrne. This figure is from Yang, X., Wang, Y., Byrne, R., Schneider, G. and Yang, S. (2019). Concepts of Artificial Intelligence for Computer-Assisted Drug Discovery. Chemical Reviews, 119(18), pp.10520–10594. doi: https://doi.org/10.1021/acs.chemrev.8b00728

Large pharmaceutical companies integrate AI-driven tools for predictive modelling to reduce R&D timelines. These AI solutions are helping companies minimise waste, cut costs, and enhance safety by predicting potential risks in chemical processes. In Midwest universities, institutions like Purdue University and University of Illinois Urbana-Champaign are pioneering AI research in chemistry, exploring machine learning techniques to predict molecular behaviour and discover new materials for energy storage and environmental applications.

Startups like Citrine Informatics, headquartered in Illinois, have gained recognition for using AI to accelerate materials innovation, collaborating with both industry and academia to enhance R&D efficiency. These developments are placing the Midwest at the forefront of AI-driven chemical innovation, creating a competitive edge in a traditionally manufacturing-heavy region.?

Research Highlights

Figure 2. “A schematic depiction of ICDC's mission”- pioneering the development of innovative supported catalysts for the next generation. Image from University of Chicago, department of chemistry.

Professor Laura Gagliardi is a theoretical and computational chemist and the Richard and Kathy Leventhal Professor of Chemistry and Molecular Engineering at the University of Chicago. Professor Gagliardi research focuses on developing quantum chemical methods to address critical challenges in renewable energy. By combining multireference theories with density functional theory, they create advanced models that offer greater accuracy in understanding complex molecular systems. They also develop force fields from first principles, which are used in classical simulations to explore the behaviour of materials at the atomic level. These methods are applied to a wide range of areas, including catalysis, carbon dioxide capture, photochemical processes, spectroscopy, and the chemistry of heavy elements. In “using nature’s blueprint to expand catalysis with Earth-abundant metals” she highlights how catalysis plays a vital role in producing modern materials, fuels, and chemicals, traditionally relying on precious metals for their high activity and stability. In contrast, metalloenzymes utilize Earth-abundant metals (EAMs), which are far more abundant and environmentally friendly. Advances in tuning the local environments around EAM active sites—similar to enzyme engineering—have led to impressive catalytic performance, sometimes surpassing that of precious metals. However, challenges remain, such as improving the activity, selectivity, and durability of EAM catalysts for industrial applications.

Figure 3. This graphic illustrates the layers of scientific challenges in computational chemistry. At the core are fundamental issues tied to the chemical elements of target systems, shaping their computational complexity. Figure provided from Rosa Di Felice et al (2023). A Perspective on Sustainable Computational Chemistry Software Development and Integration. Journal of Chemical Theory and Computation, 19(20), pp.7056–7076. doi:https://doi.org/10.1021/acs.jctc.3c00419

Professor Theresa Lynn Windus, a distinguished professor at Iowa State University and the Ames Laboratory, is a chemist specialising in high-performance computational chemistry. Her research focuses on developing advanced methods to address environmental challenges, such as creating new catalysts and renewable energy solutions. Professor Windus and her team use algorithms for high-performance computational chemistry, applying them to both basic and applied research. Her work focuses on areas like rare earth and heavy element chemistry, catalysis, aerosol formation, cellulose degradation, and photochemistry. With a strong background in software development, she has contributed to projects such as GAMESS and NWChem. Theresa also serves on the editorial board of the Journal of Physical Chemistry A/B/C and is a subject editor for Parallel Computations. In recognition of her contributions, she was named a Fellow of the American Chemical Society in 2020. There are further links to her groups work and current research here.

From her recent work she has highlighted the transformative role of quantum chemistry in predicting the ground and excited state properties of complex chemical systems, a field that has evolved through advances in theory, applied mathematics, and computer science. In this paper they explain that to fully understand the potential of these resources, the authors emphasise the need for a flexible, forward-looking strategy. They call for investments in sustainable and interoperable software infrastructure, which is essential for utilising exascale capabilities.

Figure 4. “Outline of the data acquisition.” Image from Day, A.L., Wahl, C.B., Gupta, V., Roberto dos Reis, Liao, W., Mirkin, C.A., Dravid, V.P., Choudhary, A. and Agrawal, A. (2024). Machine Learning-Enabled Image Classification for Automated Electron Microscopy. Microscopy and Microanalysis, 30(3), pp.456–465. doi: https://doi.org/10.1093/mam/ozae042

Alexandra Day and colleagues have been exploring machine learning to allow predict material properties, greatly speeding up the discovery of new materials. This is exciting development as traditionally, materials discovery relied on intuition and empirical evidence rather than systematic design. However, the rise of "big data" and significant advancements in computational power have transformed this process. For example, creation of combinatorial megalibraries, which generate millions of unique nanoparticles on a single chip and necessitate automated characterization tools. This paper introduces a specialised ML model designed for real-time binary classification of grayscale high-angle annular dark-field images of these nanoparticles. With an emphasis on minimising false positives while maintaining effectiveness on unseen data, this model tackled various computational challenges, including memory management and training optimisation using Neural Architecture Search tools. The final model exceeded expectations, achieving over 95% precision and a weighted F-score exceeding 90% on test data.

Carla Mann and colleagues from Iowa State University have looked into computational prediction of RNA-Protein interactions. In this work they explain that identifying proteins bound to specific promoter-associated RNAs (paRNAs) through experimental methods can be challenging, often requiring significant time, effort, and resources. This chapter presents a computational framework as an efficient alternative, outlining three web-based tools that can predict RNA-protein interactions. In addition, the chapter provides tables listing other useful webservers and software for RNA-protein interaction predictions, along with databases containing valuable data on known RNA-protein complexes and RNA-binding protein recognition sites.

Figure 5. “Forebrain and Spinal cord RosetteArrays cryopreserved cell banks” from Lundin, B.F., Knight, G.T., Fedorchak, N.J., Krucki, K., Iyer, N., Maher, J.E., Izban, N.R., Roberts, A., Cicero, M.R., Robinson, J.F., Iskandar, B.J., Willett, R. and Ashton, R.S. (2024). RosetteArray Platform for Quantitative High-Throughput Screening of Human Neurodevelopmental Risk. bioRxiv. doi: https://doi.org/10.1101/2024.04.01.587605 . Time-lapse images show micropatterned rosette tissue development from days 3 to 6 of FB RosetteArray formation and a plot tracks the shape of the polarised N-cadherin ring (flat, curled, or ring) to document the 2D to 3D rosette tissue morphogenesis.

Neural organoids have transformed the study of human neurodevelopmental disorders (NDDs), but their use in drug discovery and screening complex NDD causes is limited by scalability. Researchers from University of Wisconsin-Madison have collaborated with University of California and University of Chicago to analyse to put together the RosetteArray?, a platform that addresses this with a standardised, off-the-shelf 96-well plate assay that generates over 9,000 3D forebrain and spinal cord organoids from cryopreserved human stem cells in just 6-8 days. Using AI-based software, the organoids can be easily quantified. This platform is ideal for screening developmental neurotoxicity and factors linked to neural tube defects, offering a scalable solution for high-throughput neurodevelopmental risk assessment, and identifying opportunities to develop new therapeutics.

Exceptional Female Chemist

In this quarters newsletter we would like to introduce Professor Lynn Kamerlin as this month’s exceptional female chemist. Currently she is holding the prestigious position of Vasser-Woolley Georgia Research Alliance (GRA) Eminent Scholar Chair in Molecular Design. She joined Georgia Tech in 2022 after a successful position at Uppsala University in Sweden. Her research focuses on computational methods to understand and design enzymes, with significant implications in biomedicine and environmental sustainability. Professor Lynn Kamerlin began her academic career with a Master of Natural Sciences from the University of Birmingham, UK, in 2002, followed by a PhD in Theoretical Organic Chemistry in 2005. Her postdoctoral work took her to leading institutions such as the University of Vienna and the University of Southern California, where she collaborated with renowned scientists like Stefan Boresch and Arieh Warshel. She is a fellow of the Royal Society of Chemistry and serving as a Senior Editor for Protein Science. She was also awarded a European Research Council (ERC) Starting Grant and led the Young Academy of Europe in 2014-2015.

Professor Lynn Kamerlin's group studies the connection between chemistry and biology. They use computational tools to solve complex problems related to biomolecules. To understand the dynamical and mechanistic factors that influence the evolution of new enzyme functions and engineer enzymes with specific physiochemical properties tailored for various applications.

To achieve these objectives, they use advanced techniques such as replica exchange empirical valence bond simulations to analyse chemical reactivity and employ machine learning algorithms to facilitate enzyme design. They also focus on exploring natural language models to predict protein structure and activity. This approach is integrated with enhanced sampling techniques and structural bioinformatics tools to analyse conformational dynamics. For example for protein tyrosine phosphatases and quorum-quenching lactonases, which can disrupt bacterial communication. In terms of new enzymes, they are looking at protein scaffolds with catalytic functions and the incorporation of non-canonical amino acids into industrially relevant enzymes, enhancing their design and efficiency.

Opportunities and funding ?

●??????? Younger Chemists Committee Leadership Development Award. Further details found on the positions available here. Application deadlines are 15th October 2024.

●??????? The Jonathan L. Sessler Fellowship for Emerging Leaders in Bioinorganic and Medicinal Inorganic Chemistry. Application deadlines are 1st November 2024.

●??????? Heh-Won Chang, PhD Fellowship in Green Chemistry. $5,000 (Graduate students are eligible; open to U.S. & International applicants). Application deadlines are 1st November 2024.

●??????? David A. Evans Award for the Advancement and Education of Organic Synthesis. The award will consist of $5,000 and a certificate. Up to $2,500 for travel expenses to the meeting at which the award will be presented will be reimbursed. Application deadlines are 1st November 2024.

Networking events