dSPRINT

dSPRINT predicts ligand-binding sites within protein domains for DNA, RNA, ions, peptides, and small molecules to characterize ligand-binding properties across human protein domain families.

Key Features:

• Comprehensive Ligand Prediction: Predicts whether a protein domain binds DNA, RNA, small molecules, ions, or peptides and identifies specific domain positions likely involved in these interactions.
• Ensemble Machine Learning: Employs an ensemble machine learning approach to enhance predictive accuracy and robustness.
• Validation: Validated through stringent cross-validation testing, demonstrating strong predictive performance in identifying ligand-binding sites.
• Training Data: Trains models on known co-crystal structures to infer binding sites in uncharacterized domains.
• Application to DUFs: Applies predictions to Domains of Unknown Function (DUFs) to aid characterization of their molecular roles.
• Transferability to Gene Sequences: Transfers domain-level predictions to sequences, enabling extrapolation of ligand-binding properties across 95% of human genes.
• Coverage: Addresses approximately two-thirds of human protein domain families that lack detailed co-crystal structure data and provides predictions for 6,503 human protein domains.

Scientific Applications:

• Protein function annotation: Uses predicted ligand-binding sites to annotate molecular functions of protein domains.
• Characterization of DUFs: Facilitates assignment of molecular roles to domains of unknown or poorly characterized function.
• Mechanistic studies: Supports investigation of molecular mechanisms by identifying putative interaction sites within domains.
• Genome-scale functional inference: Enables broad extrapolation of ligand-binding properties across human genes to inform functional genomics analyses.

Methodology:

dSPRINT uses an ensemble machine learning approach trained on known co-crystal structures and evaluated with stringent cross-validation to predict ligand-binding sites and domain positions.