Cancerous tumors require nutrients and oxygen to grow and spread. They achieve this bypromoting the formation of new blood vessels through angiogenesis. Vascular Endothelial GrowthFactor Receptor ۲ (VEGFR۲) is central to this process. It is a receptor on the surface of cells on whichthe Vascular Endothelial Growth Factor (VEGF) is banded. When VEGF binds to VEGFR۲, it triggersa signaling cascade inside the endothelial cells. These signals stimulate the endothelial cells to multiplyand form new blood vessels. The inhibition of VEGFR۲, a key receptor in this pathway, has emergedas a critical approach to hinder tumor growth and spread by targeting the angiogenesis process essentialfor tumor sustenance and expansion. Traditional
۳D-QSAR techniques like Comparative MolecularField Analysis (CoMFA) and Comparative Molecular Similarity Index Analysis (CoMSIA) rely onclassical mechanics. Despite their success, these approaches have certain limitations, particularly whenit comes to computing the molecular electrostatic field based on atom-centered partial charges forindividual ligand molecules. To address these limitations, we present a
۳D-QSAR method for designingdrugs targeting
VEGFR۲ based on quantum mechanics. This approach offers a more precise descriptionof molecular interactions, so that it addresses some disadvantages of classical methods. We generateelectrostatic potential surfaces to gain profound insights into molecular interactions with the VEGFR۲protein at the molecular level [۱]. This information is pivotal to the precise design of effective drugs.We harnessed Density Functional Theory (DFT) calculations to obtain a Point Cloud (PC)representation as well as the Electrostatic Surface Potential (ESP) of a set of compounds. It led to adataset of PCs and ESPs related to ۱۳۰۶ drugs.
Deep learning (DL) enhances QSAR models by directlyusing raw PCs and alleviating the need for traditional feature extraction and dimension reduction. Wehave developed a unique deep learning model that uses the principles of PointNet [۲] and Transformer[۳] to predict the effectiveness of compounds on VEGFR۲. Our model demonstrated robustness andeffectiveness, achieving a Precision of ۰.۸۳۹, Recall of ۰.۸۲۴, F۱ Score of ۰.۸۲۳, Accuracy of ۰.۸۲۶,Matthews Correlation Coefficient (MCC) of ۰.۶۶۳, and an Area Under the Curve (AUC) of ۰.۸۷۳.