Session 9 – Artificial Intelligence for Solving Sustainability Problems

Sustainable development is one of key societal challenges and is, perhaps, the next big targets after addressing the more immediate climate change emergency. The two are, however, linked. Both, climate change and sustainability are system level problems: addressing any one element of the large system of technosphere does not allow to solve the problem. In order to achieve progress in addressing the climate change emergency and in establishing the foundations of sustainable technological society, we must be able to address these as system-level problems.

What are the relevant problems in the field of chemistry? The difference of Green Chemistry and Sustainable Chemistry is in the system-level analysis of the impacts of innovations and developments in chemistry in the latter. One of such system-level problems is the transition from current linear manufacturing paradigm to circular economy, when circular use of molecules will impact on all stages of chemical supply chain and development. Another important system-level change is the emergence of bio-economy.

The defining characteristic of a system-level problem is the existence of ‘emergent’ behaviour that is related to the existence of the system, and is not apparent from the individual elements of the system. Discovery, understanding and exploitation of such features is where machine learning and artificial intelligence tools are particularly suitable.

There are already a number of first example applications of AI/ML to sustainable chemistry problems. One of these is the prediction of life cycle impacts of new molecules and processes. Another, is the development of a chemical industry virtual reality, populated by AI-agents. Our group is working on the topics of feedstocks substitution and rapid process development guided by sustainability multi-objective drivers. In this talk specific examples of datascience, ML and AI techniques will be presented, using problems of finding chemical routes to target products from bio-feedstocks and developing new processes which satisfy economic and environmental objectives simultaneously.

Introduction: The purpose of the study is to use the Entellect collaborative data science environment to enable predictive retrosynthesis modelling of polymer formulations using deep learning over Reaxys reaction data to investigate the synthesis of colourant additives to enable end-of-life recycling of polymers.

Using machine learning technologies to support life-cycle-thinking within chemistry is a goal that is growing in popularity; an effective mechanism to explore bringing this into practice is through hackathon and datathon exercises. Elsevier plans to explore this approach using both the newly developed retrosynthesis tool within Reaxys, as well as using the data science environment built within the Entellect platform to allow exploration and experimentation of machine learning methods.

The project will bring teams of experts together with access to both source data on reactions and the availability of features sets ready for machine learning alongside access to resources within an environment where machine learning algorithms can be trained, tested and validated.

Elsevier has previously had success with this approach in the area of predictive analytics for drug repurposing for rare diseases with the virtual collaboration of data scientists and subject matter experts working together and is looking to replicate this success in the area predictive retrosynthesis to explore the synthesis of new molecules.

Methods: The Entellect environment makes Elsevier data available in a form that can be used to create features for machine learning allowing synthesis routes to be modelled and evaluated and allowing constraints on synthesis routes to be added into the learning for the network.

Results: The virtual datathon was intended to run in the second half of 2020 with results reported before the end of the year. However, due to the Coronavirus emergency this will be rescheduled, most likely to run in 2022.

Discussion: The study will illustrate the potential to use these approaches to investigate green and sustainable synthesis of recyclable plastics.

By the year 2030, it is estimated that the chemical production will become the main driver of worldwide oil consumption (Kätelhön et al., 2019), stressing the chemical industries’ impact on global warming. The development of sustainable process routes based on renewables streams is thus highly desirable, but challenging due to the competitiveness of the severely optimised petrochemical based market. This requires a system-wide study of the chemical supply chain to (a) identify most promising integration points for renewable feedstock (Weber et al., 2019), and to (b) compare different multi-step scenarios, identifying the most promising reaction routes. In this work we mined chemical reactions from Reaxys® database (1), which contains more than 119 million compounds and 46 million reactions, assembled a network structure by connecting reactants with their products, and applied an optimisation formalism in order to identify optimal chemical routes. We studied an industrially relevant example from β-pinene, a by-product from paper manufacturing, to citral, an important compound in fragrance and flavour industries. By using generic and simplistic sustainability indicators, we enabled a python-based rapid, easy interpretable, and customisable early-stage comparison scheme of relevant process routes. Most notably, this is a step in the direction towards competitive process development based on renewables.

References

A. Kätelhön, R. Meys, S. Deutz, S. Suh, A. Bardow, 2019, Climate change mitigation potential of carbon capture and utilization in the chemical industry, Proceedings of the National Academy of Sciences, 166, 23, 11187–11194.

J. M. Weber, P. Lió, A. A. Lapkin, 2019, Identification of strategic molecules for future circular supply chains using large reaction networks, Reaction Chemistry and Engineering, 4, 1969–1981.

(1) Copyright © 2019 Elsevier Limited except certain content provided by third parties. Reaxys is a trademark of Elsevier Limited. Rights to use Reaxys data granted via the R&D Collaboration Network.