3Cnet

3Cnet predicts the pathogenicity of human genetic variants using recurrent neural networks with multitask learning that integrates evolutionary constraints and clinical variant data.

Key Features:

• Recurrent Neural Network Modeling: Uses recurrent neural networks to analyze the amino acid sequence context surrounding human genetic variants.
• Multitask Learning Framework: Integrates evolutionary constraints and clinical pathogenicity data through multitask learning to improve variant classification performance.
• Evolutionary Conservation-Based Training: Incorporates simulated variants reflecting evolutionary conservation patterns to distinguish pathogenic from benign variants.
• Improved Pathogenicity Detection Sensitivity: Demonstrates increased sensitivity in identifying disease-causing variants compared with existing prediction approaches.

Scientific Applications:

• Clinical Variant Interpretation: Supports pathogenicity assessment of human genetic variants identified in genome sequencing studies.
• Genetic Disease Research: Enables investigation of functional effects of amino acid variants associated with human diseases.

Methodology:

The model applies recurrent neural networks to amino acid sequence contexts and trains a multitask learning framework using simulated variants representing evolutionary conservation together with clinical variant data to predict pathogenicity.