A systematic framework enhancing molecular screening efficiency in drug discovery via scaffold-driven fuzzy similarity and adaptive spectral clustering.

The drug discovery process remains slow, costly, and prone to high failure rates, with inefficiencies in lead compound identification and biological activity prediction representing significant contributing factors. Traditional QSAR models, which often focus on specific compound families and smaller datasets, struggle to meet the large-scale data requirements for deep learning models. This study proposes an innovative molecular screening and clustering method that addresses the limitations of traditional QSAR models and screening approaches. By introducing the concept of molecular scaffolds and a fuzzy molecular similarity classification method, the proposed approach enhances screening efficiency by narrowing the chemical space and provides a more nuanced classification of molecular similarity using fuzzy logic membership functions. Additionally, an adaptive Gaussian kernel function is introduced to dynamically adjust according to the dataset's characteristics, improving spectral clustering performance. The effectiveness of the proposed method in molecular screening and homologous compound identification is demonstrated through a case study using 1-naphthol as the target molecule, with all experimental data sourced from the publicly available PubChem database.