Geometricus

Geometricus represents protein structures as fixed-length vectors using shape-mers derived from 3D moment invariants for structural embedding and downstream predictive analyses.

Key Features:

• Shape-Mer Representation: Discretizes protein structure fragments into shape-mers based on 3D moment invariants and counts them to form a vector representing the full structure.
• Alignment-Free Methodology: Generates embeddings without structural alignment, enabling comparison across distantly related proteins.
• Fixed-Length Vector Embedding: Produces fixed-dimensional vectors suitable for quantitative comparison, indexing, and similarity search.
• Universal Applicability: Applicable across different protein sizes, folds, and topologies.
• Machine Learning Integration: Produces interpretable, count-based feature vectors that can be used as inputs for machine learning predictive models.
• Performance: Provides fast and accurate computation for embedding and search tasks.

Scientific Applications:

• Structure Similarity Search: Enables fast and efficient searches for similar protein structures via fixed-dimensional embeddings.
• Unsupervised Clustering: Supports clustering of proteins based on structural similarities without prior labeling.
• Structure Classification: Facilitates classification of protein structures across superfamilies and within families.
• Predictive Modeling: Supplies feature vectors for supervised machine learning models to predict protein properties.

Methodology:

Transforms protein structure fragments into shape-mers using a set of 3D moment invariants and counts shape-mer occurrences to produce fixed-length, alignment-free vector embeddings that can be used in machine learning models.