Optimizing chemistry in flow with artificial intelligence

Financiador

Organização

Publicações

Optimizing chemical reaction conditions with artificial intelligence

Publication . Almeida, A. Filipa; Rodrigues, Tiago Correia de Oliveira; Moreira, Rui Ferreira Alves; Ataíde, Filipe André Prata

The synthesis of molecules with desired properties is an important part of some areas of science and technology, including drug discovery, chemical biology, materials science, and engineering. The execution of synthetic chemistry requires expert chemical knowledge, which is acquired over several years of hands-on laboratory practice. Finding the conditions that improve the process is a time-consuming task but essential to reduce the associated cost. Artificial Intelligence (AI) and Machine learning (ML) have emerged as a problem-solving paradigm by recognizing complex patterns in data and making predictions on future data. The improvement in computer power, challenging data and algorithms provide a basis to create methods and tools to explore numerous topics/areas, such as molecule design, retrosynthetic planning, prediction of reaction outcomes, and laboratory automation. AI/ML has also seen applications in the optimization of reaction protocols. Reaction optimization is fundamental to synthetic chemistry, including the optimization of the quality, yield, or the number of steps of the process to selecting conditions for the preparation of organic compounds. AI/ML provides efficient methods to study multi-dimensional experimental spaces and accelerate the identification of the set of optimum conditions. The continuous flow technology has emerged as one of the most important techniques in green engineering. This technology offers several benefits compared to batch synthesis, including higher safety and process control that are governed by more efficient mass and heat transfer. Together, these enable the execution of reactions with hazardous components, the intensification of the process by increasing efficiency, reducing energy consumption, cost, volume and waste and direct scalability. The optimization of chemical reactions using AI/ML algorithms combined with the continuous flow can significantly reduce the overall process development time. In this work, three types of chemical reactions relevant to the pharmaceutical industry were studied: (i) Friendländer reaction (ii) Lithiation reaction, and (iii) Fluorination reaction. Due to the conditions that are required by the Lithiation and fluorinated reactions these examples were studied in continuous flow. Using tool molecules selected from the literature, it was investigated and discussed the impact of AI/ML in the optimization of reaction protocols by leveraging a bespoke random forest (RF) for reaction space exploration and synthesis protocol optimization, applied to different chemistries. To monitor and visualize the navigation of chemical space manifold learning – t-Distributed Stochastic Neighbor Embedding (t-SNE) – was used. The evaluation of the predictive power of the models was achieved by 10-fold crossvalidation and Y-randomization studies as a naïve baseline. The RF model showed better performance compared to baselines with and without regularization (linear, ridge and lasso regression). The results suggest that the RF can quickly identify optimal reaction conditions in a discretized search space by executing less than 5% of all possible reactions.

2022-09Tese de doutoramento

Acesso embargado

Ver mais

Entidade financiadora

Fundação para a Ciência e a Tecnologia

Número da atribuição

PD/BD/143125/2019