SPOT-Contact-Single

SPOT-Contact-Single predicts protein contact maps from single amino-acid sequences to support three-dimensional protein structure determination for proteins lacking homologous sequences.

Key Features:

• Input Representation: Uses outputs from the ESM-1b pre-trained language model as input features.
• Model Architecture: Integrates an ensemble of residual neural networks trained on a large dataset.
• Single-sequence Prediction: Operates without evolutionary profiles, enabling predictions for sequences with few or no homologs.
• Performance: Reports a 15% increase in F1-score versus SSCpred on the CASP14-FM test set and outperforms TrRosetta and SPOT-Contact by 48.7% and 48.5%, respectively, for Neff=1 proteins in the SPOT-2018 dataset.
• Efficiency: Provides a faster alternative to evolutionary-profile-based methods while maintaining competitive accuracy.

Scientific Applications:

• Protein Structure Prediction: Supplies contact-map constraints to assist in three-dimensional protein structure modeling.
• Functional Annotation: Extends contact-map prediction to proteins with limited sequence homology, aiding functional inference for orphan or novel sequences.

Methodology:

Uses ESM-1b outputs as input features combined with a large training set and an ensemble of residual neural networks to perform single-sequence-based contact-map prediction, removing dependency on evolutionary information.