ATHENA: A knowledge-based hybrid backpropagation-grammatical evolution neural network algorithm for discovering...

Area of Science:

• Genetics and Bioinformatics
• Computational Biology
• Human Disease Genetics

Background:

• Genetic association studies have identified limited heritability for complex traits.
• Non-additive gene-gene interactions are hypothesized to explain 'missing' heritability.
• Advanced computational methods are needed to detect these interactions.

Purpose of the Study:

• To improve the ATHENA algorithm for detecting gene-gene interactions.
• To enhance the modeling of genetic influences on human traits.
• To address the challenge of 'missing' heritability in complex diseases.

Main Methods:

• Modified ATHENA (Analysis Tool for Heritable and Environmental Network Associations) with evolutionary algorithms.
• Implemented alternative tree-based crossover and backpropagation for neural network weights.
• Integrated domain knowledge from biological databases to initialize the search.

Main Results:

• Alternative crossover modestly increased sensitivity for detecting gene-gene interactions.
• Backpropagation significantly improved performance in fitting neural network weights.
• Domain knowledge initialization led to substantial performance gains, especially in large search spaces.

Conclusions:

• Hybrid optimization, novel crossover strategies, and domain knowledge integration markedly enhance ATHENA's sensitivity.
• The improved ATHENA algorithm is more effective for detecting and modeling gene-gene interactions.
• This advancement contributes to understanding the genetic basis of complex human traits.