Deep Ensemble Learning with Atrous Spatial Pyramid Networks for Protein Secondary Structure Prediction

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Protein secondary structure is crucial for understanding 3D protein structure and function.
• Existing methods utilize models like Long Short-term Memory (LSTM) and Convolutional Neural Networks (CNN).
• Recent advancements include Gated Convolutional Neural Networks (GCNN) and Conditionally Parameterized Convolution (CondConv) in NLP and image processing.

Purpose of the Study:

• To propose a novel Conditionally Parameterized Convolutional network (CondGCNN) for improved protein secondary structure prediction.
• To leverage ensemble methods combining LSTM and CondGCNN for superior protein sequence feature extraction.
• To adapt image segmentation techniques, specifically Atrous Spatial Pyramid Pooling (ASPP), for capturing fine details in secondary structure prediction.

Main Methods:

• Developed a novel Conditionally Parameterized Convolutional network (CondGCNN).
• Employed an ensemble encoder integrating LSTM and CondGCNN for protein sequence encoding.
• Introduced an ASP network (Atrous Spatial Pyramid Pooling based network) for enhanced secondary structure detail capture.

Main Results:

• The proposed CondGCNN method achieved superior performance on protein secondary structure prediction across multiple benchmark datasets (CB513, Casp11-14).
• Ablation studies confirmed the effectiveness of individual components within the CondGCNN framework.
• The model demonstrated higher accuracy compared to existing state-of-the-art methods.

Conclusions:

• The CondGCNN approach offers a significant advancement in protein secondary structure prediction.
• The method's adaptability suggests potential applications in various protein-related prediction tasks.
• This work highlights the successful integration of advanced deep learning techniques for biological sequence analysis.