International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

Multi-Fusion Ensemble CNN for Drug–Target Binding Affinity Prediction Using Transformer-Based Molecular and Protein Representations

Authors :

Betsabeh Tanoori¹

1- Zand Institute of Higher Education

Keywords :

Drug–target binding affinity (DTA)،Convolutional neural networks،ChemBERTa،ProtBERT،Transformer embeddings،Ensemble learning

Abstract :

Accurate prediction of drug–target binding affinity (DTA) is essential in computational drug discovery, yet many deep learning models exhibit limited generalization to unseen drug–target pairs. To address this, we propose a novel Multi-Fusion Ensemble Convolutional Neural Network (MF-CNN) framework that leverages transformer-based embeddings—ChemBERTa for drug molecules and ProtBERT for protein sequences. These high-level representations are integrated using five complementary fusion strategies, each generating independent affinity predictions to capture diverse interaction perspectives. A final adaptive combination module aggregates these outputs to exploit their complementary strengths. We evaluated MF-CNN on both the widely used Davis dataset and the KIBA benchmark. Ablation studies confirm the contribution of each fusion strategy and the ensemble module. These results showcase the potential of multi-fusion ensemble architectures in advancing reliable and generalizable DTA prediction.