SSCpred: Single-Sequence-Based Protein Contact Prediction Using Deep Fully Convolutional Network.

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Protein contact prediction is crucial for understanding protein structure and function.
• Current state-of-the-art methods struggle with proteins lacking homologous sequences.
• Existing methods rely heavily on homology information, which is insufficient for non-homology targets.

Purpose of the Study:

• To develop a novel single-sequence-based contact predictor (SSCpred).
• To improve protein contact prediction accuracy for proteins with limited homology information.
• To explore the potential of deep learning for contact prediction using only the target sequence.

Main Methods:

• Developed SSCpred, a deep fully convolutional network (Deep FCN) predictor.
• Utilized pair-wise encoding of the target sequence without external homology data.
• Employed a novel pipeline leveraging Deep FCN for contact map inference.

Main Results:

• SSCpred achieves accurate protein contact predictions using only the target sequence.
• The method demonstrates competitive performance on non-homology targets compared to recent approaches.
• Experimental results validate the efficiency and accuracy of the SSCpred pipeline.

Conclusions:

• Single-sequence-based contact prediction is feasible and effective.
• SSCpred successfully compensates for the limitations of homology-dependent methods.
• The developed SSCpred pipeline offers a valuable tool for predicting contacts in proteins with low homology.