Diverse Applications of Evolutionary Computation in Bioinformatics: Hypermotifs and Gene Regulatory Network Inference

Reverse engineering genetic regulatory networks (GRNs) is greatly undetermined by the data available. We need to understand the plausibility of a recovered GRN, but little is known about the correlation between matching the target expression vector and recovery of the target GRN. Here, we explore this and related issues and claim that (i) evolved target GRNs are more reliably reconstructed by evolutionary algorithms (EAs) than are 'random' target GRNs, and (ii) there is often no correlation between the best fit expression vector and recovery of the target GRN. Put together, this suggests that EA methods for biological-GRN reverse-engineering are favoured, even if other methods more closely match the target expression vector(s). 43 Carey Pridgeon and David Corne, ‘Genetic network reverse-engineering and network size; can we identify large GRNs?’, in Computational Intelligence in Bioinformatics and Computational Biology, CIBCB '04. Proceedings of the 2004 IEEE Symposium, pages 3236, IEEE CS Press Abstract The reverse engineering of genetic regulatory networks (GRNs) is a highly challenging optimisation problem, surrounded by many unresolved questions concerning the extent to which we can regard a reverse-engineered GRN to reflect the target GRN, which we call the fidelity of the reverse engineered GRN. Related questions concern the ability of reverse-engineering algorithms to find networks that fit the data under consideration, that is, their accuracy. Most research works with networks two orders of magnitude smaller than those of biological interest, and the following question is consequently unexplored: how can we expect fidelity and accuracy to vary will reveal whether with network size? Answers to this question will reveal whether or not we can reliably extrapolate, to large networks, results obtained on the ability of reverseengineering methods on small networks. We use real-world data to explore accuracy and fidelity of a simple GRN reverse engineering approach, over sizes 100 to 6,000. We find that accurate of networks varying from networks can be found with ease at any size. However, the diversity of accurate reverse-engineered GRNs increases sharply between 100 and around 2,000 genes, then settling down to a maximal level, indicating that the fidelity of reverse-engineered networks is likely to decrease sharply with size.The reverse engineering of genetic regulatory networks (GRNs) is a highly challenging optimisation problem, surrounded by many unresolved questions concerning the extent to which we can regard a reverse-engineered GRN to reflect the target GRN, which we call the fidelity of the reverse engineered GRN. Related questions concern the ability of reverse-engineering algorithms to find networks that fit the data under consideration, that is, their accuracy. Most research works with networks two orders of magnitude smaller than those of biological interest, and the following question is consequently unexplored: how can we expect fidelity and accuracy to vary will reveal whether with network size? Answers to this question will reveal whether or not we can reliably extrapolate, to large networks, results obtained on the ability of reverseengineering methods on small networks. We use real-world data to explore accuracy and fidelity of a simple GRN reverse engineering approach, over sizes 100 to 6,000. We find that accurate of networks varying from networks can be found with ease at any size. However, the diversity of accurate reverse-engineered GRNs increases sharply between 100 and around 2,000 genes, then settling down to a maximal level, indicating that the fidelity of reverse-engineered networks is likely to decrease sharply with size. Carey Pridgeon and David Corne, ‘Hypermotifs: novel discriminatory patterns for nucleotide sequences and their application to core promoter prediction in eukaryotes’, IEEE Symposium on Computational Intelligence in Bioinformatics and Computational Biology IEEE Computational Intelligence Society, 2005.