SSCpred

SSCpred predicts protein residue contacts from single sequences using pair-wise encoding and a deep fully convolutional network (Deep FCN) to enable contact-map inference for proteins lacking homologous sequence information.

Key Features:

• Single-sequence input: Operates solely on the target sequence without requiring homologous sequence information.
• Deep FCN: Uses a deep fully convolutional network (Deep FCN) to perform contact prediction.
• Pair-wise encoding: Applies a pair-wise encoding technique to represent residue relationships for contact inference.
• Homology-independent performance: Designed to improve prediction accuracy for low-homology or nonhomologous targets.
• Reduced feature redundancy: Focuses on intrinsic single-sequence properties rather than complex or redundant feature sets.
• Contact map output: Produces residue-residue contact predictions (contact maps) for downstream structure analysis.
• Benchmark performance: Demonstrated competitive performance compared to recent methods, particularly for nonhomology targets.

Scientific Applications:

• Protein residue contact prediction: Predicts residue-residue contacts directly from single sequences.
• Contact-map inference for low-homology proteins: Infers contact maps when homologous sequence data are scarce or absent.
• Support for structure prediction: Provides contact constraints useful for protein structure prediction in challenging low-homology contexts.

Methodology:

Pair-wise encoding of the target sequence coupled with a deep fully convolutional network (Deep FCN) to predict residue-residue contacts.