GraphGONet: a self-explaining neural network encapsulating the Gene Ontology graph for phenotype prediction on gene...

Area of Science:

• Bioinformatics
• Computational Biology
• Medical Informatics

Background:

• Precision medicine leverages omics data for personalized patient care.
• Deep learning (DL) models offer advanced analytical capabilities for omics data.
• Clinical adoption of DL is hindered by the lack of interpretable predictions.

Purpose of the Study:

• To develop a knowledge-based deep learning model for explainable precision medicine.
• To integrate biological domain knowledge into a self-explaining neural network architecture.

Main Methods:

• Proposed GraphGONet, a novel self-explaining neural network.
• Encapsulated Gene Ontology within the network's hidden layers, with neurons representing biological concepts.
• Combined patient gene expression profiles with information from neighboring neurons.

Main Results:

• GraphGONet achieves state-of-the-art accuracy comparable to non-explainable models.
• The model automatically generates stable and interpretable explanations based on key biological concepts.
• These explanations facilitate clinical use by domain experts.

Conclusions:

• Knowledge-based deep learning models are a promising solution for explainable AI in medicine.
• GraphGONet provides accurate predictions with intelligible explanations, supporting clinical decision-making.
• The developed tool enhances the application of omics data in precision medicine.