DANN: a deep learning approach for annotating the pathogenicity of genetic variants

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Identifying pathogenic genetic variants, particularly non-coding ones, is crucial but challenging.
• Combined Annotation Dependent Depletion (CADD) uses support vector machines (SVMs) for variant annotation, outperforming many methods.
• SVMs have limitations in capturing non-linear feature interactions, potentially impacting annotation accuracy.

Purpose of the Study:

• To develop a novel deep learning-based algorithm for improved genetic variant annotation.
• To leverage deep neural networks (DNNs) to overcome the limitations of SVMs in capturing complex genetic feature relationships.
• To enhance the identification of pathogenic genetic variants through advanced computational methods.

Main Methods:

• Developed Deep Attentional Neural Network (DANN), a deep neural network (DNN) model.
• Utilized the same feature set and training data as the CADD algorithm.
• Employed GPU acceleration and deep learning techniques like dropout and momentum training for efficient DNN training.

Main Results:

• DANN demonstrated a 19% relative reduction in error rate compared to CADD.
• DANN achieved a 14% relative increase in the area under the curve (AUC) metric over CADD.
• The DNN architecture effectively captures non-linear relationships, improving annotation performance.

Conclusions:

• Deep learning approaches, like DANN, offer superior performance for genetic variant annotation compared to traditional machine learning methods.
• DANN provides a more accurate method for identifying pathogenic genetic variants, especially non-coding ones.
• The developed DANN algorithm and its source code are publicly available for research use.