Home > Technology peripherals > AI > body text

Combining quantum features and 20,000 molecular dynamics simulations, a new protein-ligand complex ML data set was published in the Nature sub-journal

WBOY
Release: 2024-06-01 18:20:09
Original
413 people have browsed it

Combining quantum features and 20,000 molecular dynamics simulations, a new protein-ligand complex ML data set was published in the Nature sub-journal

Editor | Dry Leaf Butterfly

Large-scale language models have greatly enhanced scientists’ ability to understand biology and chemistry, but structure-based drug discovery, quantum chemistry, and structure There are still few reliable methods in biology. Accurate biomolecule-ligand interaction datasets are urgently needed for large language models.

In order to solve this problem, researchers from the Institute of Biology of the Helmholtz Research Center München and the Technical University of Munich proposed MISATO. This is a data set that combines quantum mechanical (QM) properties of small molecules with associated molecular dynamics (MD) simulations of approximately 20,000 experimental protein-ligand complexes, and extensive validation of experimental data.

Starting from existing experimental structures, researchers systematically improved these structures using semi-empirical quantum mechanics. These include molecular dynamics simulations of a large number of protein-ligand complexes in pure water, with accumulation times exceeding 170 microseconds.

The team provides an example of a machine learning (ML) baseline model demonstrating improved accuracy by using this dataset. Provides machine learning experts with an easy entry point to implement next-generation artificial intelligence models for drug discovery.

The study is titled "MISATO: machine learning dataset of protein–ligand complexes for structure-based drug discovery" and was published in "Nature" on May 10, 2024. Computational Science》.

Combining quantum features and 20,000 molecular dynamics simulations, a new protein-ligand complex ML data set was published in the Nature sub-journal

In recent years, AI prediction technology has triggered a revolution in the scientific field. For example, AlphaFold can accurately predict protein structure. Although structure-guided drug discovery remains a huge challenge, the application of AI in this field is still shallow. Current methods face challenges such as accuracy, computational cost, and experimental dependence, and mostly focus on simple solutions and one-dimensional data processing. The complexity of three-dimensional protein-ligand complexes has been overlooked.

Although a variety of databases exist, no AI model has been shown to advance drug discovery due to limitations in data volume and lack of thermodynamic information. Unlike AlphaFold's achievements in the field of protein structure prediction, the AI ​​model is also limited by ignoring issues such as dynamics and chemical complexity, which affects its potential in biomolecule analysis and quantum chemistry.

Here, researchers from the Institute of Structural Biology of the Helmholtz Research Center München and the Technical University of Munich proposed a protein-ligand structure database based on experimental protein-ligand structures, MISATO (Molecular Interactions Are Structurally Optimized).

Researchers have shown that the database can help better train models in areas related to drug discovery and beyond. This includes quantum chemistry, general structural biology and bioinformatics.

Combining quantum features and 20,000 molecular dynamics simulations, a new protein-ligand complex ML data set was published in the Nature sub-journal

Illustration: MISATO combines QM data with MD-derived protein ligand dynamics. (Source: paper)

The team provides quantum chemistry-based structure management and refinement, including regularization of ligand geometries. The researchers augmented this database with missing dynamic and chemical information, including MD on time scales, allowing the detection of transient and mysterious states of certain systems. The latter is very important for successful drug design.

Combining quantum features and 20,000 molecular dynamics simulations, a new protein-ligand complex ML data set was published in the Nature sub-journal

Illustration: PDBbind database optimized according to quantum chemistry protocols. (Source: paper)

Therefore, the researchers supplemented the experimental data with the maximum number of physical parameters. This relieves the AI ​​model from the burden of learning all this information implicitly, allowing it to focus on the main learning task. The MISATO database provides a user-friendly format that can be imported directly into machine learning code.

Combining quantum features and 20,000 molecular dynamics simulations, a new protein-ligand complex ML data set was published in the Nature sub-journal

Illustration: Experimental validation of QM, MD and AI models. (Source: paper)

The team also provides various preprocessing scripts to filter and visualize the dataset. Furthermore, example AI baseline models are provided for calculating quantum chemical properties (chemical hardness and electron affinity), binding affinity calculations, and predicting protein flexibility or induced fit characteristics, allowing the data to be simplified. Moreover, QM, MD, and AI models have been extensively validated on experimental data.

The researchers hope to transform MISATO into a beneficial community project that will benefit the entire field of drug discovery.

Paper link:https://www.nature.com/articles/s43588-024-00627-2

The above is the detailed content of Combining quantum features and 20,000 molecular dynamics simulations, a new protein-ligand complex ML data set was published in the Nature sub-journal. For more information, please follow other related articles on the PHP Chinese website!

Related labels:
source:jiqizhixin.com
Statement of this Website
The content of this article is voluntarily contributed by netizens, and the copyright belongs to the original author. This site does not assume corresponding legal responsibility. If you find any content suspected of plagiarism or infringement, please contact admin@php.cn
Popular Tutorials
More>
Latest Downloads
More>
Web Effects
Website Source Code
Website Materials
Front End Template
About us Disclaimer Sitemap
php.cn:Public welfare online PHP training,Help PHP learners grow quickly!