(492b) Machine Learning-Driven, High-Throughput Solvent Screening Technique for Biobased 2,3-Butanediol Extraction

Biobased 2,3-Butanediol (2,3-BDO) has gained tremendous attention due to its capability of being transformed into a variety of useful products across the chemical, food, pharmaceutical, aerospace, and manufacturing industries [1]. 2,3-BDO can be produced via microbial fermentation, which can decrease the reliance on nonrenewable, petroleum-based chemicals, and thus result in a cheaper, simpler, and more environmentally friendly process. Despite its potential economic and environmental advantages, the major bottleneck of 2,3-BDO’s bioproduction is the separation and purification process [2]. Researchers have investigated the use of solvent extraction as an efficient and cost-effective method for 2,3-BDO separation [3]. However, due to 2,3-BDO’s high boiling point, high hydrophilicity, and low concentration, the optimal solvent or solvent mixture for 2,3-BDO extraction is still under investigation. To bridge this gap, the integration of machine learning (ML) and autonomous high-throughput (HT) technologies provides a unique opportunity to screen various solvent types more efficiently than conventional experimentation [4].

In this talk, we explore the implementation of an ML-driven, HT workflow as a solvent screening method to optimize the separation of 2,3-BDO in bioprocesses. A dataset of 3459 entries for BDO’s solubility in organic solvents was curated based on publicly accessible databases. It allowed for the development of a Histogram Gradient Boosting model to accurately predict 2,3-BDO’s solubility in organic solvents. An initial list of 55 solvent candidates was generated based on the screening of the ML model. A narrowed set of 24 solvent candidates was selected for experimental investigation. The influence of experimental parameters such as solvent feed ratio, equilibrium time, and temperature, on the solvent distribution coefficient and extraction efficiency of 2,3-BDO were evaluated. Furthermore, the development of an automated liquid handling platform for HT solvent screening is established. To this end, the present study demonstrates the use of an ML-driven, automated HT workflow as a rapid solvent screening method for the optimized solvent extraction of 2,3-BDO. Such findings lay the foundation for the use of AI-driven technologies for quicker and more accurate screening of solvents for a wide range of bio-separation applications.

[1] S. Xie, Z. Li, G. Zhu, W. Song, and C. Yi, “Cleaner production and downstream processing of bio-based 2,3-butanediol: A review,” Journal of Cleaner Production, vol. 343. Elsevier Ltd, Apr. 01, 2022. doi: 10.1016/j.jclepro.2022.131033.

[2] T. Rajale et al., “Separation, recovery and upgrading of 2,3-butanediol from fermentation broth,” Biofuels, Bioproducts and Biorefining, vol. 17, no. 4, pp. 1003–1011, Jul. 2023, doi: 10.1002/bbb.2496.

[3] Q. W. Zhang, L. G. Lin, and W. C. Ye, “Techniques for extraction and isolation of natural products: A comprehensive review,” Chinese Medicine (United Kingdom), vol. 13, no. 1. BioMed Central Ltd., Apr. 17, 2018. doi: 10.1186/s13020-018-0177-x.

[4] D. Liu and N. Sun, “Prospects of artificial intelligence in the development of sustainable separation processes,” Frontiers in Sustainability, vol. 4, 2023, doi: 10.3389/frsus.2023.1210209.