(546c) Use of Octanol-Water Partition Coefficient and Toxicity Prediction in Design of Ionic Liquids for in Situ Fermentation Extraction

Fermentation is widely used to produce fine and bulk chemicals Due to toxicity of the product towards the producing microorganisms, overall product concentrations are often dilute. After fermentation, downstream separation is required which often has high energy inputs. In situ product removal with an immiscible solvent, such as certain ionic liquids (ILs),offers a mechanism by which product could be continually removed during production, reducing the effects of toxicity and product inhibition while also decreasing downstream separation demands. The IL solvent should have good partitioning of the product, but must have low toxicity to the microorganism to allow in situ processing.  Environmental and health concerns should be addressed to ensure sustainability. Ionic liquids, consisting of cations and anions, are highly tunable towards fermentation extractions.  For instance, they can be designed with hydrophobicity, low to no vapor pressure, low toxicity, etc.  Hundreds of ionic liquids have been synthesized with up to 10¹⁴ possible combinations. The use of molecular descriptors within property prediction models provides an efficient method of searching the possible chemical space and selecting/designing candidate ionic liquids for a given extraction.

Predictive models for octanol-water partition coefficient and toxicity for various model organisms (Escherichia coli, Vibrio Fischeri, and rat) are presented. Experimental data collected from previously published literature and presented here have been used to develop the models. Ionic liquid descriptors were calculated using E-Dragon (http://www.vcclab.org/lab/edragon/). Descriptor selection was performed using an exhaustive search to find descriptors that provided a good fit to the experimental data (high R²) while avoiding overfitting (represented by Mallow’s C_p statistic). Using the selected descriptors, linear regression was used to generate the predictive property models. Leave-one-out cross-validation (LOOCV) was used to evaluate the predictive power of the models. Once the models were finalized, they were used in a computational molecular design (CMD) framework to optimally design ionic liquids for a given extraction. The optimization problem was formulated as a mixed-integer nonlinear program (MINLP) where a molecule structure is determined which minimizes the difference between target properties and the predicted properties of the solution molecule. A stochastic method, Tabu search, was used to identify locally optimal solutions to the optimization problem. The overall framework has been used previously by the authors in the design of ionic liquids for use in refrigeration. The target properties included low octanol-water partition coefficient, low E. coli toxicity, low V. Fischeri toxicity (indicative of aqueous environmental safety) and low rat toxicity (indicative of human toxicity). The results indicate that several designed ionic liquid candidates have desirable properties for solvent extraction from fermentation reactions. Further property prediction tools will allow more refined design of ionic liquids for more specified fermentation applications.