Multiplex Molecular Graph Neural Network (MXMNet): An Overview
The use of artificial intelligence (AI) in drug discovery is becoming increasingly popular. One approach to this problem is to use a technique called representation learning where a machine learning model learns the features or characteristics of a molecule based on its structure, function, and interactions. MXMNet is one such approach for representation learning that focuses on the interactions between molecules.
The Construction of a Multiplex Graph
The interactions between molecules are divided into two categories: local and global. The local layer only contains the local connections that mainly capture covalent interactions. On the other hand, the global layer contains the global connections that cover non-covalent interactions. These interactions form the basis of a two-layer multiplex graph represented as G = {Gl, Gg} where Gl and Gg denote the local and global layers, respectively. The construction of the multiplex graph allows the machine learning model to differentiate between the types of interactions between molecules.
The MXM Module
The MXM module used in MXMNet contains two types of message passing. The first type of message passing is an angle-aware message passing operated on Gl. The angle-aware message passing takes into account the direction of the bonds between atoms in a molecule while passing messages between them. The second type of message passing is an efficient message passing operated on Gg. This is because the global layer usually contains a larger number of interactions and requires an efficient mechanism for message passing.
The combination of these two types of message passing allows the machine learning model to learn the interactions between molecules better.
Benefits of MXMNet
The use of MXMNet has several benefits in drug discovery. Firstly, it allows scientists to predict the properties of a molecule before it is synthesized, saving time and resources. Secondly, it can be used to identify potential drug targets and lead compounds that can be further optimized through chemical modifications. Thirdly, it can be used to identify potential off-target effects and toxicity of a drug.
In summary, MXMNet is an approach for the representation learning of molecules that relies on a two-layer multiplex graph and a novel angle-aware message passing mechanism. The use of MXMNet has several benefits in drug discovery, including predicting the properties of a molecule, identifying potential drug targets and lead compounds, and identifying potential off-target effects and toxicity of a drug.
