همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Generating Pharmaceutical Molecules Using Multi-Path Deep Learning

نویسندگان :

Sanaz Hashemipour¹ Habib Izadkhah² Abolfazl Barzegar³

1- Dept. of Computer Science, Faculty of Mathematics, Statistics, and Computer Science, University of Tabriz 2- Associate Professor, Dept. of Computer Science, Faculty of Mathematics, Statistics, and Computer Science, University of Tabriz 3- Associate Professor, Research Center of Biosciences & Biotechnology (RCBB), University of Tabriz

کلمات کلیدی :

multi-path LSTM (MP-LSTM)،de novo drug design،deep learning،bioinformatics،artificial intelligence in medicinal chemistry

چکیده :

The growing complexity of molecular data and the vast chemical space, encompassing an extraordinarily large number of potential compounds, pose significant challenges in drug discovery. In this work, the Multi-Path Long Short-Term Memory (MP-LSTM) model is proposed, a hybrid deep learning architecture for generating pharmaceutical molecules. The model employs multiple learning pathways that integrate LSTM and one-dimensional convolutional layers, enabling the extraction of diverse features from input sequences. Experimental results demonstrate that the MP-LSTM outperforms conventional LSTM and Bi-LSTM models in producing novel and chemically valid drug-like molecules. The generated compounds were further evaluated using the SwissADME platform and the RDKit library, confirming favorable pharmacokinetic properties. These results highlight the efficacy of the MP-LSTM model in exploring chemical space and demonstrate its potential to accelerate AI-driven drug discovery.