Automatic classification of protein structures using low-dimensional structure space mappings.

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Protein structure dictates protein function, making structure-function relationships crucial in biochemistry.
• Manual protein classification databases like SCOP (Structural Classification of Proteins) are high-quality but laborious and subjective.
• The rapid increase in solved protein structures necessitates automated classification methods.

Purpose of the Study:

• To develop an automated method for protein structure classification.
• To create low-dimensional maps of protein structure space (MPSS) for classification.
• To achieve automated classification comparable to manual standards like SCOP at higher hierarchical levels.

Main Methods:

• Constructing MPSS using pairwise alignment distances from CE, Dali, FATCAT, and MATT.
• Employing multidimensional scaling (MDS) for low-dimensional embedding.
• Determining an optimal distance threshold using SCOP's Superfamily and Fold levels to link map adjacency to functional relationships.

Main Results:

• MPSS accurately represent protein fold space, enabling automated classification.
• The automated classification closely matches SCOP's Superfamily and Fold levels.
• This method surpasses previous automated approaches, which typically only match the Family level.

Conclusions:

• MPSS provide a robust framework for automated protein structure classification.
• The approach successfully classifies remote protein homologies at Superfamily and Fold levels.
• MDS offers a superior noise-reducing transformation for protein structure distances compared to existing methods.