Limits...
Optimization of neural network architecture using genetic programming improves detection and modeling of gene-gene interactions in studies of human diseases.

Ritchie MD, White BC, Parker JS, Hahn LW, Moore JH - BMC Bioinformatics (2003)

Bottom Line: Appropriate definition of neural network architecture prior to data analysis is crucial for successful data mining.Using simulated data, we show that a genetic programming optimized neural network approach is able to model gene-gene interactions as well as a traditional back propagation neural network.Furthermore, the genetic programming optimized neural network is better than the traditional back propagation neural network approach in terms of predictive ability and power to detect gene-gene interactions when non-functional polymorphisms are present.

View Article: PubMed Central - HTML - PubMed

Affiliation: Program in Human Genetics and Department of Molecular Physiology and Biophysics, Vanderbilt University Medical School, Nashville, TN, 37232-0700, USA. ritchie@phg.mc.vanderbilt.edu

ABSTRACT

Background: Appropriate definition of neural network architecture prior to data analysis is crucial for successful data mining. This can be challenging when the underlying model of the data is unknown. The goal of this study was to determine whether optimizing neural network architecture using genetic programming as a machine learning strategy would improve the ability of neural networks to model and detect nonlinear interactions among genes in studies of common human diseases.

Results: Using simulated data, we show that a genetic programming optimized neural network approach is able to model gene-gene interactions as well as a traditional back propagation neural network. Furthermore, the genetic programming optimized neural network is better than the traditional back propagation neural network approach in terms of predictive ability and power to detect gene-gene interactions when non-functional polymorphisms are present.

Conclusion: This study suggests that a machine learning strategy for optimizing neural network architecture may be preferable to traditional trial-and-error approaches for the identification and characterization of gene-gene interactions in common, complex human diseases.

Show MeSH
Optimal architecture from BPNN trial and error optimization. Thisfigure shows the result of the BPNN trial and error procedure onone data set from each epistasis model. This shows the NN architecturefor the best classification error selected from Table 1.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC183838&req=5

Figure 5: Optimal architecture from BPNN trial and error optimization. Thisfigure shows the result of the BPNN trial and error procedure onone data set from each epistasis model. This shows the NN architecturefor the best classification error selected from Table 1.

Mentions: The results of our trial-and-error optimization technique forselecting the traditional BPNN architecture indicate the importanceof selecting the optimal NN architecture for each different epistasismodel. Table 1 shows the resultsfrom the traditional BPNN architecture optimization technique onone data set from each epistasis model generated with the two functionalSNPs only. A schematic for each of the best architectures selectedis shown in Figure 5. Note thatthe optimal architecture varied for each of the epistasis models.For example, the optimal architecture for Model 1 was composed of2 hidden layers, 15 nodes in the first layer and 5 nodes in thesecond layer, and a momentum of 0.9. In contrast, for Model 2 theoptimal architecture included 2 hidden layers, 20 nodes in the firstlayer and 5 nodes in the second layer, and a momentum of 0.9. Differentarchitectures were optimal for models 3, 4, and 5 as well. Similarresults were obtained using the simulated data containing the twofunctional and eight non-functional SNPs as well (data not shown).This provides further motivation for automating the optimizationof the NN architecture in order to avoid the uncertainty of trial-and-errorexperiments.


Optimization of neural network architecture using genetic programming improves detection and modeling of gene-gene interactions in studies of human diseases.

Ritchie MD, White BC, Parker JS, Hahn LW, Moore JH - BMC Bioinformatics (2003)

Optimal architecture from BPNN trial and error optimization. Thisfigure shows the result of the BPNN trial and error procedure onone data set from each epistasis model. This shows the NN architecturefor the best classification error selected from Table 1.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC183838&req=5

Figure 5: Optimal architecture from BPNN trial and error optimization. Thisfigure shows the result of the BPNN trial and error procedure onone data set from each epistasis model. This shows the NN architecturefor the best classification error selected from Table 1.
Mentions: The results of our trial-and-error optimization technique forselecting the traditional BPNN architecture indicate the importanceof selecting the optimal NN architecture for each different epistasismodel. Table 1 shows the resultsfrom the traditional BPNN architecture optimization technique onone data set from each epistasis model generated with the two functionalSNPs only. A schematic for each of the best architectures selectedis shown in Figure 5. Note thatthe optimal architecture varied for each of the epistasis models.For example, the optimal architecture for Model 1 was composed of2 hidden layers, 15 nodes in the first layer and 5 nodes in thesecond layer, and a momentum of 0.9. In contrast, for Model 2 theoptimal architecture included 2 hidden layers, 20 nodes in the firstlayer and 5 nodes in the second layer, and a momentum of 0.9. Differentarchitectures were optimal for models 3, 4, and 5 as well. Similarresults were obtained using the simulated data containing the twofunctional and eight non-functional SNPs as well (data not shown).This provides further motivation for automating the optimizationof the NN architecture in order to avoid the uncertainty of trial-and-errorexperiments.

Bottom Line: Appropriate definition of neural network architecture prior to data analysis is crucial for successful data mining.Using simulated data, we show that a genetic programming optimized neural network approach is able to model gene-gene interactions as well as a traditional back propagation neural network.Furthermore, the genetic programming optimized neural network is better than the traditional back propagation neural network approach in terms of predictive ability and power to detect gene-gene interactions when non-functional polymorphisms are present.

View Article: PubMed Central - HTML - PubMed

Affiliation: Program in Human Genetics and Department of Molecular Physiology and Biophysics, Vanderbilt University Medical School, Nashville, TN, 37232-0700, USA. ritchie@phg.mc.vanderbilt.edu

ABSTRACT

Background: Appropriate definition of neural network architecture prior to data analysis is crucial for successful data mining. This can be challenging when the underlying model of the data is unknown. The goal of this study was to determine whether optimizing neural network architecture using genetic programming as a machine learning strategy would improve the ability of neural networks to model and detect nonlinear interactions among genes in studies of common human diseases.

Results: Using simulated data, we show that a genetic programming optimized neural network approach is able to model gene-gene interactions as well as a traditional back propagation neural network. Furthermore, the genetic programming optimized neural network is better than the traditional back propagation neural network approach in terms of predictive ability and power to detect gene-gene interactions when non-functional polymorphisms are present.

Conclusion: This study suggests that a machine learning strategy for optimizing neural network architecture may be preferable to traditional trial-and-error approaches for the identification and characterization of gene-gene interactions in common, complex human diseases.

Show MeSH