Armadillo: domain boundary prediction by amino acid composition

Area of Science:

• Structural biology
• Bioinformatics
• Computational biology

Background:

• Protein domain identification is crucial for determining molecular function.
• Large, multi-domain proteins and flexible linkers pose challenges for experimental and computational structure determination.
• Existing domain prediction methods struggle with poorly conserved or orphan proteins due to reliance on sequence similarity.

Purpose of the Study:

• To present a simple computational method for identifying protein domain linkers and their boundaries from sequence data.
• To introduce the Armadillo domain predictor and its underlying amino acid index, the domain linker propensity index (DLI).

Main Methods:

• Converted protein sequences into smoothed numeric profiles using amino acid indices.
• Derived the domain linker propensity index (DLI) from the amino acid composition of known domain linkers.
• Predicted domain linker boundaries using Z-score distributions of the numeric profiles.

Main Results:

• The DLI index revealed a propensity for Pro and Gly in linker residues.
• Armadillo achieved 35% sensitivity with DLI alone on a two-domain, single-linker dataset.
• Combining DLI with an entropy-based index increased sensitivity to 56% for two-domain proteins.
• Armadillo demonstrated 37% sensitivity for multi-domain proteins, outperforming existing methods.

Conclusions:

• Armadillo offers a simple yet effective method for predicting protein domain boundaries with reasonable sensitivity.
• The tool is valuable for analyzing poorly conserved or orphan proteins, aiding in structural and functional annotation.
• Armadillo predictions can enhance the performance of domain meta-predictors.