FISH Amyloid

FISH Amyloid identifies amyloidogenic segments in protein sequences by analyzing site-specific amino acid co-occurrence patterns to predict regions prone to amyloid formation.

Key Features:

• Machine learning-based classification: The method employs an original machine learning algorithm that classifies amino acid sequences by identifying discriminative patterns of site-specific co-occurrences.
• Co-occurrence matrix analysis: The tool computes co-occurrence matrices from positive (amyloidogenic) and negative (non-amyloidogenic) training sequences and uses matrix comparisons to inform classification.
• Matrix distance measurement: Classification leverages maximal distances between co-occurrence matrices derived from positive and negative instances.
• Training and performance: The method was trained on datasets of hexapeptides with known amyloidogenic properties and reported AUC up to 0.80 for experimental data and 0.95 for computationally generated datasets.
• Versatility in segment length: Trained on sequences ranging from 4 to 10 residues, the approach handles short peptides and longer protein segments such as Sup35 prion protein regions.
• Sliding window approach: A sliding window of specified length scans protein sequences to detect discriminative co-occurrence patterns.

Scientific Applications:

• Neurodegenerative disease research: Identification of amyloidogenic regions supports studies of diseases associated with misfolded proteins such as Alzheimer's and Parkinson's.
• Disease mechanism and target identification: Predicted amyloid-prone segments can inform investigation of pathogenic aggregation mechanisms and potential therapeutic targets.
• Peptide and protein screening: The method applies to screening short peptides and longer protein sequences, including analysis of prion proteins like Sup35.

Methodology:

The algorithm identifies the most relevant training segment in each positive instance via co-occurrence pattern analysis, computes and compares co-occurrence matrices for positive and negative instances measuring maximal distances for classification, scans sequences with a sliding window of specified length, and was trained on datasets of hexapeptides with known amyloidogenic properties.