Gated Recurrent Unit with Attention Mechanism for IC50 Prediction Model Using Amyotrophic Lateral Sclerosis Related Proteins

Amyotrophic Lateral Sclerosis is a common neurodegenerative disease that weakens the brain and spinal cord myelin nerves. Two protein targets of Amyotrophic Lateral Sclerosis are used in this study. The protein targets selected are SOD1 and TBK1. Numerous SOD1 mutations in the genome have been linked to ALS. The inhibition of the harmful effects of the defective SOD1 protein aids in ALS treatment. ALS has been linked to variations in the TBK1 gene. Modulating the immune system and neural inflammation is achieved through inhibition of TBK1. IC50 prediction is an important strategy for selecting drugs during Drug discovery. In this paper, Deep Learning models are built using a Gated Recurrent Unit (GRU) that accepts SMILES (Simplified Molecular Input Line Entry System) representation of active molecules as input. The key novelty in the proposed research is a customized GRU IC50 prediction model with attention mechanism. A new GRU model with an attention mechanism is implemented. Drug molecules and their features are collected from the ChEMBL database. More than 500 ligands are used and converted to SMILES one hot encoding for training the models. The prediction results of models are found using important metrics and observed more accurate predictions of IC50.