James Alexander Hughes

Benchmark datasets for the DNA fragment assembly problem

Many computational intelligence approaches have been used for the fragment assembly problem. However, the comparison and analysis of these approaches is difficult due to the lack of availability of standard benchmarks. Although similar datasets may be used as a starting point, there is not enough information to reproduce the exact overlaps matrix for the fragments used by the various approaches, creating a problem for consistency. This paper presents a collection of benchmark datasets for a wide range of fragment lengths, number of fragments, and sequence lengths, along with a description of the method used to produce them. A website has been created to maintain the datasets and the tables of results at http://chac.sis.uia.mx/fragbench/. Researchers are invited to add to the datasets by following the method described, as well as to submit results obtained by their algorithms on the benchmarks.

Recentering, reanchoring & restarting an evolutionary algorithm

Recentering-restarting evolutionary algorithms have been used successfully to evolve epidemic networks. This study develops multiple variations of this algorithm for the purpose of evaluating its use for ordered-gene problems. These variations are called recentering or reanchoring-restarting evolutionary algorithms. Two different adaptive representations were explored that both use generating sets to produce local search operations. The degree of locality is controllable by setting program parameters. The variations and representations are applied to what may be considered the quintessential ordered gene problem, the Travelling Salesman Problem. Two sets of experimental analysis were performed. The first used large problem instances to determine how well this algorithm performs in comparison to benchmarks obtained from the DIMACS TSP implementation challenge. The second used many small problem instances to determine if any one of the recentering/reanchoring-restarting evolutionary algorithms outperforms the others. Variations of the recentering/reanchoring-restarting evolutionary algorithm were comparable to some of the best performing computational intelligence algorithms. In studying the small problem instances, no significant trend was found to suggest that one variation of baseline evolutionary algorithms or recentering/reanchoring-restarting evolutionary algorithms outperformed the others. This study shows that the new algorithms are very useful tools for improving results produced by other heuristics.

Edit metric decoding: Representation strikes back

Quaternary error-correcting codes defined over the edit metric may be used as labels to track the origin of sequence data. When used in such applications there are typically additional restrictions that are biologically motivated, such as a required GC content or the avoidance of certain patterns. As a result such codes can not be expected to have a regular structure, making decoding particularly challenging. Previous work on decoding edit codes considered the use of side effect machines for decoding, successfully decoding up to 93.86% of error vectors. In this study the recentering/restarting algorithm is used in combination with side effect machines and an alternative representation based upon transpositions. Using the same data as in the previous work, the rate of successful decoding was significantly improved, with many cases obtaining rates very close to 100%.

Recentering and restarting a genetic algorithm using a generative representation for an ordered gene problem

Recentering-Restarting Genetic Algorithms have been used successfully to evolve multiple epidemic networks and perform DNA error correction. This work studies variations of the Recentering-Restarting Genetic Algorithm for the purpose of evaluating its effectiveness for ordered gene problems. These variations use multiple seeds and two adaptive representations which use generating sets to produce local search. These algorithm variations are applied to what many considered the quintessential ordered gene problem, the Travelling Salesman Problem. Two distinct sets of experimental analysis was performed: first, using large problem instances to determine the effectiveness of the Recentering-Restarting Genetic Algorithm in comparison to benchmarks and second, studying many small problem instances ranging from 12 to 20 cities to determine if any one of the algorithm variations always outperforms the others.These algorithm variations were comparable to highly competitive optimization algorithms submitted to the DIMACS TSP implementation challenge. In studying the small problem instances, it was observed that no one algorithm always dominates on all problem instances within a domain. This study demonstrates how the Recentering-Restarting Genetic Algorithm is a useful tool for improving upon results generated by other powerful heuristics.

Recentering and Restarting Genetic Algorithm variations for DNA Fragment Assembly

The Fragment Assembly Problem is a major component of the DNA sequencing process that is identified as being NP-Hard. A variety of approaches to this problem have been used, including overlap-layout-consensus, de Bruijn graphs, and greedy graph based algorithms. The overlap-layout-consensus approach is one of the more popular strategies which has been studied on a collection of heuristics and metaheuristics. In this study heuristics and Genetic Algorithm variations are combined to exploit their respective benefits. These algorithms were able to produce results that surpassed the best results obtained by a collection of state-of-the-art metaheuristics on ten of sixteen popular benchmark data sets.

Restarting and recentering genetic algorithm variations for DNA fragment assembly: The necessity of a multi-strategy approach

DNA Fragment assembly - an NP-Hard problem - is one of the major steps in of DNA sequencing. Multiple strategies have been used for this problem, including greedy graph-based algorithms, deBruijn graphs, and the overlap-layout-consensus approach. This study focuses on the overlap-layout-consensus approach. Heuristics and computational intelligence methods are combined to exploit their respective benefits. These algorithm combinations were able to produce high quality results surpassing the best results obtained by a number of competitive algorithms specially designed and tuned for this problem on thirteen of sixteen popular benchmarks. This work also reinforces the necessity of using multiple search strategies as it is clearly observed that algorithm performance is dependent on problem instance; without a deeper look into many searches, top solutions could be missed entirely.

Finding Nonlinear Relationships in fMRI Time Series with Symbolic Regression

The brain is an intrinsically nonlinear system, yet the dominant methods used to generate network models of functional connectivity from fMRI data use linear methods. Although these approaches have been used successfully, they are limited in that they can find only linear relations within a system we know to be nonlinear. This study employs a highly specialized genetic programming system which incorporates multiple enhancements to perform symbolic regression, a type of regression analysis that searches for declarative mathematical expressions to describe relationships in observed data. Publicly available fMRI data from the Human Connectome Project were segmented into meaningful regions of interest and highly nonlinear mathematical expressions describing functional connectivity were generated. These nonlinear expressions exceed the explanatory power of traditional linear models and allow for more accurate investigation of the underlying physiological connectivities.

Searching for nonlinear relationships in fMRI data with symbolic regression

The vast majority of methods employed in the analysis of functional Magnetic Resonance Imaging (fMRI) produce exclusively linear models; however, it is clear that linear models cannot fully describe a system with the observed behavioral complexity of the human brain --- an intrinsically nonlinear system. By using tools embracing the possibility of modeling the underlying nonlinear system we may uncover meaningful undiscovered relationships which further our understanding of the brain. We employ genetic programming, an artificial intelligence technique, to perform symbolic regression for the discovery of nonlinear models better suited to capturing the complexities of a high dimensional dynamic system: the human brain. fMRI data for multiple subjects performing different tasks were segmented into regions of interest and nonlinear models were generated which effectively described the system succinctly. The nonlinear models contained undiscovered relationships and selected different sets of regions of interest than traditional tools, which leads to more accurate understanding of the functional networks.

Modelling intracranial pressure with noninvasive physiological measures

Patients who suffered a traumatic brain injury (TBI) require special care, and physicians often monitor intercranial pressure (ICP) as it can greatly aid in management. Although monitoring ICP can be critical, it requires neurosurgery, which presents additional significant risk. Monitoring ICP also aids in clinical situations beyond TBI, however the risk of neurosurgery can prevent physicians from gathering the data. The need for surgery may be eliminated if ICP could be accurately inferred using noninvasive physiological measures. Genetic programming (GP) and linear regression were used to develop nonlinear and linear mathematical models describing the relationships between intercranial pressure and a collection of physiological measurements from noninvasive instruments. Nonlinear models of ICP were generated that not only fit the subjects they were trained on, but generalized well across other subjects. The nonlinear models were analysed and provided insight into the studied underlying system which led to the creation of additional models. The new models were developed with a refined search, and were more accurate and general. It was also found that the relations between the features could be explained effectively with a simple linear model after GP refined the search.

An algebraic generalization for graph and tensor-based neural networks

Despite significant effort, there is currently no formal or de facto standard framework or format for constructing, representing, or manipulating general neural networks. In computational neuroscience, there have been some attempts to formalize connectionist notations and generative operations for neural networks, including Connection Set Algebra, but none are truly formal or general. In computational intelligence (CI), though the use of linear algebra and tensor-based models are widespread, graph-based frameworks are also popular and there is a lack of tools supporting the transfer of information between systems. To address these gaps, we exploited existing results about the connection between linear and relation algebras to define a concise, formal algebraic framework that generalizes graph and tensor-based neural networks. For simplicity and compatibility, this framework is purposefully defined as a minimal extension to linear algebra. We demonstrate the merits of this approach first by defining new operations for network composition along with proofs of their most important properties. An implementation of the algebraic framework is presented and applied to create an instance of an artificial neural network that is compatible with both graph and tensor based CI frameworks. The result is an algebraic framework for neural networks that generalizes the formats used in at least two systems, together with an example implementation.