Biologically Informed Neural Networks Predict Drug Responses.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Deep neural networks (DNNs) offer high accuracy in biological predictions but lack interpretability.
• Understanding the mechanisms behind DNN predictions is crucial for biological and clinical applications.

Purpose of the Study:

• To develop an interpretable deep neural network model for predicting cancer drug sensitivity.
• To leverage the Gene Ontology (GO) to structure the DNN hierarchically, enhancing model explainability.

Main Methods:

• Utilized deep neural networks with a hierarchical architecture.
• Incorporated Gene Ontology to guide the network's structure.
• Modeled the sensitivity of various cancers to a range of drugs.

Main Results:

• The hierarchical DNN successfully predicted cancer drug sensitivity.
• The model's structure, derived from GO, provided insights into the biological basis of drug response.
• Achieved high predictive accuracy while offering a degree of contextualization for predictions.

Conclusions:

• Hierarchical DNNs structured by Gene Ontology can improve the interpretability of predictive models in cancer research.
• This approach facilitates a better understanding of cancer drug sensitivity mechanisms.
• Offers a promising direction for developing explainable AI in precision oncology.