The Effect of Using Machine Learning Algorithms on the Accuracy of the Prediction of Dopamine Neurotransmitters

Overview

This project attempted to test different types of machine learning models and algorithms to identify which model was most effective for predicting the accuracy of dopamine neurotransmitters. D2 dopamine receptors were chosen because of their implication in serious conditions like Parkinson’s, schizophrenia, and substance use. However, D2 is hard to elucidate receptor-specific functions because of its degree of similarity with other proteins. A recent study was conducted on QSAR models to predict D2 and D3 bindings, as well as the selectivity for D3. This project aimed to continue this research by looking at D2 molecules and testing various machine learning algorithms, as in the study. We predicted that the machine learning model GNN (Graph Neural Network) would work better at predicting dopamine neurotransmitter activity than QSAR (Quantitative Structure-Activity Relationship) modeling and MLP (Multi Layered Perception) To study this, bioactivity data for the Dopamine D2 receptor were obtained from the ChEMBL and processed using RDKit. Molecular descriptors and fingerprints were generated for traditional QSAR and neural network models, while molecular graphs were created for the GNN. A QSAR model, a feedforward neural network, and a GNN were then trained and evaluated using ROC-AUC and precision-recall metrics. The results show that the GNN model performed significantly better than QSAR and slightly better than MLP. Furthermore, the QSAR data predicted the lipophilicity of D2 membranes accurately by showing its selectivity with hydrophobic models. This project is extremely important because many studies across the world utilize different types of models for predicting neurotransmitter activity in order to create new drug treatments for serious conditions. However, a study comparing these methods is a recent, emerging scientific question. These findings could be used for a better understanding of neurotransmitter activity and eventually be utilized to find potential drug treatments.