PredinID

PredinID predicts pathogenic inframe insertion/deletion (indel) variants in humans using graph-based machine learning to assess variant pathogenicity.

Key Features:

• Graph Convolutional Network (GCN): Uses a GCN framework to perform prediction of variant pathogenicity.
• Node Classification Modeling: Models the prediction task explicitly as a node classification problem within a feature graph.
• Feature Graph Construction (k-nearest neighbor): Constructs a feature graph using the k-nearest neighbor algorithm to aggregate informative representations.
• Edge-Based Sampling Strategy: Employs an edge-based sampling strategy to extract information from feature-space connections and subgraph topologies.
• Performance Evaluation: Evaluated with 5-fold cross-validation and tested on independent test sets, with comparisons to four classic machine learning algorithms and two other GCN methods.

Scientific Applications:

• Pathogenicity Prediction: Predicts the pathogenicity of inframe indels in human protein-coding regions.
• Genetic Disease Research: Supports investigation of the genetic basis of diseases associated with inframe indels and can inform diagnostic and therapeutic research.

Methodology:

Constructs a feature graph via k-nearest neighbor, applies a GCN for node classification, uses an edge-based sampling strategy to enhance information extraction, and evaluates performance using 5-fold cross-validation and independent test sets with comparisons to four classic machine learning algorithms and two other GCN methods.