ArchPred

ArchPred predicts protein loop structures by searching a loop-fragment library to model conformations of variable loop regions in protein structures.

Key Features:

• Fragment-search based methodology: Searches a regularly updated loop library ("Search Space") for candidate fragments matching query length and bracing secondary-structure types.
• Geometrical constraints: Imposes geometrical restraints using stem residues to filter candidate loops that fit the query geometry.
• Side-chain rebuilding and fit assessment: Rebuilds side chains on candidate loops and assesses fit by calculating root mean square deviation (r.m.s.d.) of stem regions and evaluating rigid-body clashes with the environment.
• Ranking and scoring: Ranks candidates using a Z-score that integrates sequence similarity and comparisons of predicted versus observed main-chain dihedral-angle propensities.
• Conjugate gradient minimization: Inserts the selected loop conformation and refines its placement by conjugate gradient minimization to reduce clashes and optimize geometry.

Scientific Applications:

• Loop conformation prediction: Predicts loop structures in proteins lacking experimental loop coordinates.
• Structural biology: Supports modeling of protein structures and conformational analysis of loop regions.
• Protein engineering: Assists design and modification of loops for altered function or stability.
• Drug design: Provides loop models relevant to ligand binding sites and structure-based drug design.
• Functional interpretation: Aids understanding of protein function at the molecular level through loop modeling.

Methodology:

Searches a loop library (Search Space) selecting fragments by length and bracing secondary-structure types, applies geometrical restraints on stem residues, rebuilds side chains and assesses fit via stem-region r.m.s.d. and rigid-body clash evaluation, ranks candidates by a Z-score combining sequence similarity and main-chain dihedral-angle propensity comparisons, inserts the chosen loop and refines it by conjugate gradient minimization; benchmarking used artificially prepared search datasets with trivial sequence similarities removed at the SCOP superfamily level, reporting coverage of 98%, 78%, and 28% with r.m.s.d. accuracies of at least 0.22 Å, 1.38 Å, and 2.47 Å for loop lengths 4, 8, and 12 respectively, and showing approximately a 5:1 improvement over an earlier database search method.