ABLE: Attention based learning for enzyme classification.

Area of Science:

• Bioinformatics
• Computational Biology
• Enzymology

Background:

• Accurate protein classification is crucial for biological research.
• The Protein Data Bank (PDB) offers a vast resource of protein structures.
• Existing methods for enzyme classification have limitations.

Purpose of the Study:

• To develop a novel deep learning model for classifying proteins into enzyme classes.
• To utilize only the primary protein structure (FASTA sequence) for classification.
• To compare the proposed model's performance against established methods.

Main Methods:

• An attention-based bidirectional-LSTM (ABLE) model was developed.
• The model was trained on oversampled data using SMOTE (Synthetic Minority Over-sampling Technique).
• Performance was evaluated against eighteen other machine learning and deep learning networks, including CNN, LSTM, Bi-LSTM, GRU, and DeepEC.

Main Results:

• The ABLE model achieved a highest F1-score of 0.834 on a PDB dataset.
• The model demonstrated superior performance compared to baseline models.
• Experiments included comparisons with SMOTE and ADASYN oversampling techniques.

Conclusions:

• The proposed ABLE model is effective for enzyme class prediction using protein primary structure.
• Deep learning approaches, particularly attention-based LSTMs, show promise in protein classification.
• Further research can explore advanced oversampling techniques and model architectures.