An Explainable Deep Learning Model for Clathrin Protein Prediction Using a DCT-Enhanced Position-Specific Scoring...

Area of Science:

• Computational biology
• Bioinformatics
• Protein structure and function prediction

Background:

• Clathrin protein (CP) is crucial for cellular processes like endocytosis and signal transduction.
• CP dysfunction is linked to diseases such as neurodegenerative disorders and cancer.
• Accurate CP prediction is vital for biomedical research and disease understanding.

Purpose of the Study:

• To develop a novel computational framework, Pred-CLGRUs, for predicting Clathrin protein.
• To enhance feature extraction and reduce noise using Position-Specific Scoring Matrix (PSSM) and Discrete Cosine Transform (DCT).
• To evaluate the performance of Gated Recurrent Units (GRUs) within the Pred-CLGRUs framework.

Main Methods:

• Utilized Position-Specific Scoring Matrix (PSSM) for evolutionary information and Discrete Cosine Transform (DCT) for feature enhancement.
• Employed Gated Recurrent Units (GRUs) as the primary deep learning architecture.
• Trained and compared GRUs against Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and Deep Neural Networks (DNNs).

Main Results:

• The Pred-CLGRUs framework with GRUs achieved 93.33% accuracy.
• Achieved a Matthews Correlation Coefficient of 0.86, indicating strong predictive performance.
• Demonstrated high sensitivity (94.44%) and specificity (92.22%) in Clathrin protein prediction.

Conclusions:

• The PSSM-DCT approach effectively enhances feature extraction for CP prediction.
• The Pred-CLGRUs model shows significant potential for accurate Clathrin protein prediction.
• Future work includes integrating heterogeneous biological data and exploring advanced models like Capsule Neural Networks (CapsNet).