Pafig

Pafig predicts amyloid fibril-forming hexapeptides and assesses their aggregation propensity for analysis of protein aggregation in amyloid diseases.

Key Features:

• Support Vector Machine Framework: Pafig employs support vector machines (SVMs) to classify hexapeptides by their propensity to form fibrillar aggregates.
• Physicochemical Property Utilization: Pafig uses 41 physicochemical properties sourced from the AAindex database to represent hexapeptide characteristics.
• Two-Round Feature Selection: Relevant physicochemical attributes were selected through a two-round selection process from AAindex to refine the input feature set.
• Validation and Performance Metrics: Pafig was evaluated by 10-fold cross-validation on the Hexpepset dataset, achieving 81% overall accuracy and a Matthews correlation coefficient (MCC) of 0.63.

Scientific Applications:

• Amyloid disease research: Identification of aggregation-prone hexapeptides enables mapping of potential fibril-forming regions within proteins to study molecular mechanisms of amyloid diseases.
• Therapeutic target identification and proteome screening: Proteome-wide prediction of amyloidogenic hexapeptides supports discovery of candidate regions for therapeutic intervention and large-scale proteomic analyses.

Methodology:

Pafig analyzes all possible hexapeptides (≈64 million) using an SVM model built on 41 AAindex physicochemical properties selected via a two-round process; it predicts ~5.08% of hexapeptides as likely fibril-forming, was validated by 10-fold cross-validation on the Hexpepset dataset (81% accuracy, MCC 0.63), and reports higher frequencies of alanine, phenylalanine, isoleucine, leucine, and valine and lower frequencies of aspartic acid, glutamic acid, histidine, lysine, arginine, and proline in predicted regions.