DeepDBP

DeepDBP predicts DNA-binding proteins from protein sequences using deep learning to enable computational identification of protein–DNA interactions.

Key Features:

• DeepDBP-ANN: Employs an artificial neural network trained on generated features derived from protein sequences and achieved a training accuracy of 99.02% and test accuracy of 82.80%.
• DeepDBP-CNN: Uses a pre-learned embedding combined with convolutional neural networks to capture spatial hierarchies in protein sequences and achieved a training accuracy of 94.32% and test accuracy of 84.31%, outperforming DeepDBP-ANN on test data.
• Generated sequence-derived features: Uses features generated from protein sequences for supervised ANN training.
• Pre-learned embedding and CNN input: Incorporates a pre-learned embedding as input to convolutional neural networks to enable automatic capture of spatial patterns in sequences.
• Benchmark evaluation: Both methodologies were evaluated on standard benchmark datasets and reported state-of-the-art performance in predicting DNA-binding proteins.

Scientific Applications:

• Identification of DNA-binding proteins: Facilitates high-accuracy prediction of DNA-binding proteins from sequence data for genomics and proteomics studies.
• Study of protein–DNA interactions: Enables large-scale exploration of protein–DNA interactions and their roles in genetic regulation and cellular mechanisms.
• Investigation of disease pathology: Supports studies of disease mechanisms in which DNA-binding proteins are implicated.
• Functional studies: Assists investigations into biological processes involving DNA-binding proteins, including immune defense and oxygen transport.

Methodology:

Supervised learning using artificial neural networks (DeepDBP-ANN) on generated sequence-derived features and convolutional neural networks (DeepDBP-CNN) using a pre-learned embedding, with automatic feature extraction from protein sequences.