图表基于神经网络的分子性质预测与补丁聚合
Teng Jiek See1, Daokun Zhang2, Mario Boley3
1Medicinal Chemistry, Monash Institute of Pharmaceutical Sciences, Monash University, 381 Royal Parade, Parkville, VIC 3068, Australia.
Journal of chemical theory and computation
|October 2, 2024
概括
补丁聚合是图形神经网络 (GNN) 的新方法,提高了分子性质预测准确性和参数效率. 这种新的方法可以提高计算化学预测,而不会增加模型的复杂性.
科学领域:
- 计算化学是一种计算化学.
- 机器学习是机器学习.
- 量子力学就是量子力学.
背景情况:
- 图形神经网络 (GNN) 是有效的分子属性预测.
- 当前的GNN聚合方法增加参数和计算成本,但没有保证准确度的提高.
研究的目的:
- 为GNN引入一种新的,参数效率高的边缘到节点聚合机制.
- 提高GNN在预测分子性质方面的准确性和效率.
主要方法:
- 开发了"补丁聚合",灵感来自多头注意力和专家混合.
- 在最先进的 GNN 模型 (SchNet,DimeNet++,SphereNet,TensorNet,VisNet) 中集成补丁聚合.
- 与现有方法 (sum,MLP,softmax,设置变压器) 进行补丁聚合的比较.
主要成果:
- 补丁聚合在预测QM9热力学特性和MD17能量/力方面始终优于现有的聚合技术.
- 该方法证明了预测准确度和参数效率的提高.
- 补丁聚合被证明在各种GNN架构中适用.
结论:
- 补丁聚合是GNN在分子性质预测中的优越边缘到节点机制.
- 它提供了更高的准确性和计算效率,适用于资源有限的应用程序.
- 这种方法代表了GNN在计算化学中的重大进步.
相关概念视频
Patch Clamp
5.4K
Many fundamental cell functions such as muscle contraction and nerve transmission rely on the electrical signals produced by the movement of positively and negatively charged ions across the cell membrane. One competent method to record current flowing across the whole cell or single ion channel is the patch-clamp technique.
In this method, a glass micropipette containing electrolyte solution is tightly sealed against a small portion of the cell membrane. As a result, a patch of the cell...
In this method, a glass micropipette containing electrolyte solution is tightly sealed against a small portion of the cell membrane. As a result, a patch of the cell...
5.4K
Protein Networks
3.9K
An organism can have thousands of different proteins, and these proteins must cooperate to ensure the health of an organism. Proteins bind to other proteins and form complexes to carry out their functions. Many proteins interact with multiple other proteins creating a complex network of protein interactions.
These interactions can be represented through maps depicting protein-protein interaction networks, represented as nodes and edges. Nodes are circles that are representative of a protein,...
These interactions can be represented through maps depicting protein-protein interaction networks, represented as nodes and edges. Nodes are circles that are representative of a protein,...
3.9K
Molecular Models
38.0K
Physical models representing molecular architectures of chemical compounds play essential roles in understanding chemistry. The use of molecular models makes it easier to visualize the structures and shapes of atoms and molecules.
38.0K
Protein-protein Interfaces
12.5K
Many proteins form complexes to carry out their functions, making protein-protein interactions (PPIs) essential for an organism's survival. Most PPIs are stabilized by numerous weak noncovalent chemical forces. The physical shape of the interfaces determines the way two proteins interact. Many globular proteins have closely-matching shapes on their surfaces, which form a large number of weak bonds. Additionally, many PPIs occur between two helices or between a surface cleft and a...
12.5K
Ligand Binding Sites
12.8K
Proteins are dynamic macromolecules that carry out a wide variety of essential processes; however, the activities of most proteins depend on their interactions with other molecules or ions, known as ligands.
Protein-ligand interactions are quite specific; even though numerous potential ligands surround a cellular protein at any given time, only a particular ligand can bind to that protein. Moreover, a ligand binds only to a dedicated area on the surface of the protein, known as the...
Protein-ligand interactions are quite specific; even though numerous potential ligands surround a cellular protein at any given time, only a particular ligand can bind to that protein. Moreover, a ligand binds only to a dedicated area on the surface of the protein, known as the...
12.8K


