LEGO-CSM

LEGO-CSM predicts protein functional properties, including subcellular localization, Enzyme Commission (EC) numbers, and Gene Ontology (GO) terms, by combining sequence and structural information using a structure-based Cutoff Scanning Matrix.

Key Features:

• Integration of Sequence and Structure Information: Uses protein sequence and structural data represented as graph-based signatures to capture features relevant to function.
• Structure-based Cutoff Scanning Matrix: Applies a Cutoff Scanning Matrix that leverages structural context to inform functional prediction.
• Supervised Learning Models: Constructs supervised learning models to predict labels corresponding to Subcellular Localization, Enzyme Commission (EC) numbers, and Gene Ontology (GO) terms.
• Predictive Performance: Reports AUC-ROC values up to 0.93 for subcellular localization and EC number prediction and up to 0.81 for GO term prediction in independent blind tests.

Scientific Applications:

• Proteomics: Functional annotation of newly discovered or uncharacterized proteins.
• Systems Biology: Improved assignment of protein functions to support modeling of cellular processes.
• Drug Discovery: Functional characterization of proteins to identify or prioritize potential drug targets.

Methodology:

Employs a structure-based Cutoff Scanning Matrix, integrates graph-based signatures derived from sequence and structural data, and builds supervised learning models; performance was assessed using independent blind tests with AUC-ROC metrics.