Molecular Dynamics–Guided Generative Design of Enzyme Modulators to Improve Feed Efficiency in Cattle

Abstract

Cattle feed efficiency is one of the key goals of minimizing the economic costs and environmental effects of livestock production. The existing approaches to dietary intervention using exogenous enzymes and universal-action additives are constrained by variable effectiveness, lack of stability, and off-target toxicity in the intricate rumen ecosystem whereas conventional discovery approaches to enzyme modulators are tedious and inferior to the sensitivity of targeting particular metabolic routes. This paper suggestions a new computational pipeline which combines Molecular Dynamics simulations with Generative Deep Learning in the de novo design of small-molecule enzyme modulators to inhibit important proteins in nitrogen and energy metabolism. Its methodology involves the use of enhanced-sampling MD simulations to identify cryptic and allosteric binding sites and the subsequent training of a 3D Conditional Variational Autoencoder on the conformational ensembles and in silico screening of new compounds to identify those with binding free energy, molecular docking, MM-GBSA binding free energy, and ADMET filtering. The pipeline is able to screen 10,000 candidate molecules with ten lead compounds being identified with regards to binding affinities that are predicted to be much stronger than those of reference inhibitors. The best urease inhibitor has a docking score of -11.2 kcal/mol and MM-GBSA binding free energy of -45.2 kcal/mol, which is equivalent to a 3.7-fold increase over acetohydroxamic acid. The results have proven a proof-of-concept that MD-guided generative design is a potent paradigm to develop precision livestock feed additives with possible uses in sustainable animal farming.