Biomedical and Translational Informatics Laboratory

Grammatical Evolution Neural Networks

Method Description:

GENN – Applying grammatical evolution to optimize neural nets for detection and modeling of gene-gene interactions.

Genetically Programmed Neural Networks (GPNN) is a technique that utilizes genetic programming to optimize neural nets for classification and identification of gene-gene interactions.

Grammatical evolution (GE) is an evolutionary algorithm that uses linear genomes and grammars to define the populations. In GE, each individual consists of a binary genome divided into codons. Mutation takes place on individual bits but crossover only takes place between the codons. An individual or phenotype is produced by translating the codons using the grammar. The resulting individual can then be tested for fitness in the population and the usual evolutionary operators can be carried out. By using a grammar to define the phenotype, GE separates the genotype from the phenotype and allows greater genetic diversity within the population than other evolutionary algorithms.

Since GENN uses a grammar to define the structure of the resulting neural network, we can easily vary the behavior of the program with changes to the grammar. In GPNN the GP was constrained so that only valid neural networks can be produced. Any change to the behavior required changes to the code. The constraints for GENN are provided by the grammar itself and can be easily modified without modification of the code. For example, Boolean operators can be added or removed by changing only the grammar file used as an input to the program.

In addition, GPNN uses a binary tree for the genome and therefore, only two connections between nodes are possible. In GENN the grammar allows for defining multiple connections between nodes selected by the algorithm. Variable numbers of connections allows for more complicated neural networks to be evolved and potentially makes GENN more powerful than GPNN.


  • Motsinger AA, Ritchie MD. Neural networks for genetic epidemiology: past, present, and future. BMC BioData Mining, 1:3 (2008).
  • Ritchie MD, Bartlett J, Bush WS, Edwards TL, Motsinger AA, Torstenson ES. Exploring epistasis in candidate genes for Rheumatoid Arthritis. BMC Proceedings, 1 Suppl 1:S70 (2007).
  • Motsinger-Reif AA, Fanelli TJ, Davis AC, Ritchie MD. Power of grammatical evolution neural networks to detect gene-gene interactions in the presence of error. BMC Research Notes, 1:65 (2008).
  • Motsinger-Reif AA, Dudek SM, Hahn LW, Ritchie MD. Comparison of approaches for machine learning optimization of neural networks for detecting gene-gene interactions in genetic epidemiology. Genetic Epidemiology, 32(4):325-40 (2008).
  • Motsinger-Reif AA, Reif DM, Fanelli TJ, Ritchie MD. A comparison of analytical methods for genetic association studies. Genetic Epidemiology, 32(8):767-78 (2008).
  • Motsinger AA, Reif DM, Dudek SM, Ritchie MD. Understanding the Evolutionary Process of Grammatical Evolution Neural Networks for Feature Selection in Genetic Epidemiology. IEEE Symposium on Computational Intelligence in Bioinformatics and Computational Biology, 1-8 (2006).
  • Motsinger AA, Reif DM, Fanelli TJ, Davis AC, Ritchie MD. Linkage disequilibrium in genetic association studies improves the performance of grammatical evolution neural networks. IEEE Symposium on Computational Intelligence in Bioinformatics and Computational Biology, 1-8 (2007).
  • Turner SD, Ritchie MD, Bush WS. Conquering the Needle-in-a-Haystack: How Correlated Input Variables Beneficially Alter the Fitness Landscape for Neural Networks. Lect Notes Comput Sci, 5483:80-91 (2009).
  • Turner SD, Dudek SM, Ritchie MD. Grammatical Evolution of Neural Networks for Discovering Epistasis among Quantitative Trait Loci. Lect Notes Comput Sci. 6023:86-97. (2010).
  • Motsinger AA, Dudek SM, Hahn LW, Ritchie MD. Comparison of neural network optimization approaches for studies of human genetics. Lecture Notes in Computer Science, 3907:103-114 (2006).

Related Links:

GE –