# Automated parametrization of small molecules within the Martini 3 coarse-grained model guided by experimental log P values

## TL;DR

This paper introduces an automated method to create accurate coarse-grained models of small molecules using experimental log P values, improving simulations for drug discovery.

## Contribution

A novel automated parametrization approach for small molecules in Martini 3 using a mixed-variable particle swarm algorithm and experimental log P values.

## Key findings

- The method successfully matches experimental log P values and reproduces atomistic density profiles in lipid bilayers.
- Automated parametrization improves the accuracy of coarse-grained models for drug discovery applications.
- The fitness function effectively evaluates model accuracy by combining structural and dynamic targets.

## Abstract

Molecular dynamics simulations play an important role in investigating biological systems. However, simulating large-scale systems can be computationally expensive, which can be improved by the employment of a coarse-graining force field. This study focuses on the automated parametrization of small molecules within the CGCompiler framework. This optimization approach utilizes a mixed-variable particle swarm algorithm to avoid the manual tweaking of parameters. Particularly, the optimization focuses on matching experimentally known log P values of partitioning in water-octanol phases, reproducing atomistic density profiles in lipid bilayers, and optimizing overall shape and volume aspects of the modeled atomistic molecules. After the atomistic to coarse-grained mapping, the model’s accuracy is evaluated through a fitness function, which combines structural and dynamic targets, to accurately capture the shape and behavior of the small molecule in question. Through the investigation of the interactions between small molecules and cellular membranes, this optimization process supports the development of accurate coarse-grained models for small molecules relevant to drug discovery. Our work demonstrates promising results in automating the high-fidelity parametrization of small molecules using the Martini 3 force-field guided by experimental log P values.