SPOT-1D-Single

SPOT-1D-Single predicts protein secondary structure, backbone dihedral angles, solvent accessibility, and half-sphere exposures from single amino-acid sequences without requiring evolutionary information.

Key Features:

• Predicted features: Predicts three-state (SS3) and eight-state (SS8) secondary structure, backbone dihedral angles, relative solvent accessibility, and half-sphere exposures from single-sequence input.
• Single-Sequence Methodology: Operates effectively on single-sequence inputs and does not rely on multiple sequence alignments or evolutionary profiles.
• Large Training Dataset: Trained on a dataset of 39,120 proteins deposited before 2016.
• Ensemble Learning Approach: Employs an ensemble that integrates hybrid long-short-term-memory (LSTM) bidirectional neural networks with convolutional neural networks (CNNs).
• Performance Metrics: Achieved three-state secondary structure accuracy of 72.12%–74.28% on independent test sets (TEST2018, SPOT-2016, SPOT-2016-HQ, SPOT-2018, SPOT-2018-HQ, CASP12, CASP13) and outperforms SPIDER3-Single and ProteinUnet on these sets.
• Enhanced Accuracy for Homolog-Limited Proteins: Improves SS3 and SS8 accuracies by 6.24% and 6.98% respectively relative to SPOT-1D for proteins with Neff = 1 (no homologs).

Scientific Applications:

• Protein Structure Prediction: Provides residue-level structural annotations that support inference of three-dimensional conformations from sequence.
• Functional Annotation: Supplies structural feature predictions useful for annotating unknown proteins and inferring potential functional roles.
• Drug Discovery and Design: Supplies structural information such as secondary structure and accessibility that can inform identification of potential binding sites and conformations relevant to drug-target interactions.

Methodology:

An ensemble model combining hybrid bidirectional LSTM networks and convolutional neural networks was trained on 39,120 proteins deposited before 2016 to predict secondary structure, backbone angles, solvent accessibility, and half-sphere exposures.