abstract.tex

\begin{my_abstract}

MLP models present a compelling alternative to traditional computational methods, offering significantly faster performance than ab initio quantum mechanics (QM) methods and higher accuracy than classical force fields, making them particularly suitable for molecular dynamics (MD) simulations of large biochemical systems.
In this dissertation, several developments of machine learning potential (MLP) models tailored for biochemical systems are presented.
\Cref{ch:dprc} introduces the Deep Potential Range Correction (DPRc) model to construct efficient and accurate potential energy models for free energy profiles of biochemical reactions in solution.
\Cref{ch:qdp} and \cref{ch:semiempirical} propose the Quantum Deep Potential (QD$\pi$) model particularly for drug discovery applications and show its high accuracy.
\Cref{ch:deepmd-kit} details the development of DeePMD-kit (version 2), a software suite designed for constructing and deploying MLP models.
\Cref{ch:dataset} introduces the QD$\pi$ dataset, comprising high-precision potential energy data for small organic molecules.
Finally, \cref{ch:gnn} presents DeePMD-GNN, a software package enabling the integration of graph neural network (GNN)-based models, within the DeePMD-kit framework.
Taken together, this work advances the state-of-the-art in computational chemistry, offering novel methodologies and tools for accurate and efficient modeling of complex biochemical systems.

\end{my_abstract}