Abstract A memetic approach to protein structure refinement was defined, combining Differential Evolution and the widely used Rosetta Relax refinement protocol. This refinement process can be considered as an optimization problem to optimize the positions of the amino acid atoms of the protein. The local optimization procedures of Rosetta Relax are integrated into the evolutionary algorithm. The results with different proteins show that the memetic algorithm better samples the energy landscape compared to Rosetta Relax, obtaining better energy-optimized refined conformations in the same runtime.

In this work we have experimented with the use of genetic programming as a feature selection method as well as a classifier to obtain business failure prediction models with different prediction temporal horizons. In the prediction models, a wide set of explanatory variables has been used, all of them based on the annual accounts of the company. In addition, an extended set of explanatory variables incorporating variables from the economic environment has been considered. Comparison of the prediction results between these alternatives shows a trend towards better results using the feature selection process, while there is no trend towards better results using economic environment variables.

This study focuses on dimensionality reduction by variable selection in business failure prediction models. A new method of dimensionality reduction by variable selection using Genetic Programming is proposed, which takes into account the relative frequency of occurrence of the explanatory variables in the evolved solutions, as well as the statistical relevance of that frequency. For a better evaluation of the proposed method and its comparison with other well-tested and widely used variable selection methods, the prediction of business failure in three temporal horizons (1, 5 and 9 years prior to failure) is considered. Additionally, a comparison of the sets of variables selected with different feature selection methods is performed, also considering different classifiers in the comparison, among which Genetic Programming is included as a classifier. The results indicate that the proposed method (using Genetic Programming as a variable selection method) is superior to the most tested and widely used methods analyzed, and this superiority increases if Genetic Programming is also used as a classification method.

Abstract This study uses different feature selection methods in the field of business failure prediction and tests the capability of Genetic Programming (GP) as an appropriate classifier in this field. The prediction models categorize the insolvency/non-insolvency of a firm one year in advance from a large set of financial ratios. Different selection strategies based on two evolutionary algorithms were used to reduce the dimensionality of the financial features considered. The first method considers the combination between the global search provided by an evolutionary algorithm (differential evolution) with a simple classifier, together with the possible use of classical filters in a first step of feature selection. Secondly, genetic programming is used as a feature selector. In addition, these selection approaches will be compared when GP is used exclusively as a classifier. The results show that, when using GP as a classifier method, the proposed selection method with GP stands out from the rest. Moreover, the use of GP as a classifier improves the results with respect to other classifier methods. This shows an added value to the use of GP in this field, in addition to the interpretability of GP prediction models.

In this paper we discuss the advantages and problems of two alternatives for ab initio protein structure prediction. On one hand, recent approaches based on deep learning, which have significantly improved prediction results for a wide variety of proteins, are discussed. On the other hand, methods based on protein conformational energy minimization and with different search strategies are analyzed. In this latter case, our methods based on a memetic combination between differential evolution and the fragment replacement technique are included, incorporating also the possibility of niching in the evolutionary search. Different proteins have been used to analyze the pros and cons in both approaches, proposing possibilities of integration of both alternatives.

Amino acid properties were optimized for protein secondary structure prediction. The artificial properties were evolved using differential evolution in a property space of arbitrary dimensionality. These properties were optimized to provide the correct results in predicting the elements of the protein secondary structure using a simple classifier model and with standard benchmark sets. A comparison is performed with respect to the use of the commonly employed orthogonal and neutral encoding of the amino acids of the protein chain, together with a discussion of the similarities of the evolved artificial properties with respect to physical properties of amino acids.

Computational structural biology is an important, growing research area that includes diverse problems, such as protein structure prediction, computational molecular assembly, and computer-assisted drug design problems, that include protein-ligand binding, protein-protein, protein-RNA, and protein-DNA docking, as well as computer assisted wet-laboratory structure resolving problems, like registration, reconstruction, and refinement. The focus of this special section is on the application of non-linear optimization to problems in structural biology, thus turning the spotlight on a growing area of interdisciplinary research that brings together expertise in meta-heuristic optimization and computational structural biology. The five articles included in this section provide a glimpse of the diversity of work in this area, highlighting the adaptation and use of a variety of state-of-the-art meta-heuristics for a range of problems linked to the wider area of structural biology.

BACKGROUND: The canonical code, although prevailing in complex genomes, is not universal. It was shown the canonical genetic code superior robustness compared to random codes, but it is not clearly determined how it evolved towards its current form. The error minimization theory considers the minimization of point mutation adverse effect as the main selection factor in the evolution of the code. We have used simulated evolution in a computer to search for optimized codes, which helps to obtain information about the optimization level of the canonical code in its evolution. A genetic algorithm searches for efficient codes in a fitness landscape that corresponds with the adaptability of possible hypothetical genetic codes. The lower the effects of errors or mutations in the codon bases of a hypothetical code, the more efficient or optimal is that code. The inclusion of the fitness sharing technique in the evolutionary algorithm allows the extent to which the canonical genetic code is in an area corresponding to a deep local minimum to be easily determined, even in the high dimensional spaces considered. RESULTS: The analyses show that the canonical code is not in a deep local minimum and that the fitness landscape is not a multimodal fitness landscape with deep and separated peaks. Moreover, the canonical code is clearly far away from the areas of higher fitness in the landscape. CONCLUSIONS: Given the non-presence of deep local minima in the landscape, although the code could evolve and different forces could shape its structure, the fitness landscape nature considered in the error minimization theory does not explain why the canonical code ended its evolution in a location which is not an area of a localized deep minimum of the huge fitness landscape.

We used evolutionary computing for optimizing cancer treatments taking into account the presence and effects of cancer stem cells. We used a cellular automaton to model tumor growth at cellular level, based on the presence of the main cancer hallmarks in the cells. The cellular automaton allows the study of the emergent behavior of the multicellular system evolution in different scenarios defined by the predominance of the different hallmarks. When cancer stem cells (CSCs) are modeled, the multicellular system evolution is additionally dependent on the CSC tumor regrowth capability because their differentiation to non-stem cancer cells. When a standard treatment is applied against non-stem (differentiated) cancer cells, different effects are present depending on the strategy used to eliminate these non-stem cancer cells. We used Differential Evolution to optimize the treatment application strategy in terms of intensity, duration and periodicity to minimize the final outcome of tumor growth and regrowth.

We used a hybrid combination of evolution and learning for automatically obtaining robot controllers. Additionally, we employed the complementary reinforcement backpropagation algorithm, which integrates either positive reinforcements or punishments with supervised connectionist learning for artificial neural network robot behavior controllers. The algorithm was adapted to consider a continuous range in the outputs of the neural network controller. Furthermore, we added Differential Evolution to integrate the advantages of run-time learning with those of evolutionary learning. We ran some tests for validating this approach to obtain robust robotic behavior controllers.

This paper combines the center-crossing condition in artificial neural networks that incorporate synaptic delays in their connections and which act as Central Pattern Generators (CPGs) for biped controllers. Recurrent synaptic delay based neural networks allow greater time reasoning capabilities in the neural controllers, outperforming the results of continuous time recurrent neural networks, the neural model most used as CPG for biped robot locomotion related behaviors. Simulated evolution is used to automatically obtain neural controllers for walking behaviors, showing the capabilities of the synaptic delay based neural networks for the temporal coordination of the biped joints in difficult surfaces.

We used cellular automata (CA) for the modeling of the temporal folding of proteins. Unlike the focus of the vast research already done on the direct prediction of the final folded conformations, we will model the temporal and dynamic folding process. The CA model defines how the amino acids interact through time to obtain a folded conformation. We employed the TIP model to represent the protein conformations in a lattice, we extended the classical CA models using artificial neural networks for their implementation, and we used evolutionary computing to automatically obtain the models by means of Differential Evolution. Moreover, the modeling of the folding provides the final protein conformation.

A new segmentation method using deformable models was defined for medical image segmentation. As deformable model we used Topological Active Nets, model which integrates features of region-based and boundary-based segmentation techniques. The model deformation is controlled by an Artificial Neural Network (ANN) that learns how to move the nodes of the model based on their energy surrounding. The ANN is applied to each of the nodes and in different temporal steps until the final segmentation is reached. The ANN training is obtained by simulated evolution, using Differential Evolution to automatically obtain the ANN that provides the emergent segmentation. The methodology was adapted and tested in two different medical domains, that is, CT medical images and eye fundus images to demonstrate the potential of the segmentation technique.

BACKGROUND: As the canonical code is not universal, different theories about its origin and organization have appeared. The optimization or level of adaptation of the canonical genetic code was measured taking into account the harmful consequences resulting from point mutations leading to the replacement of one amino acid for another. There are two basic theories to measure the level of optimization: the statistical approach, which compares the canonical genetic code with many randomly generated alternative ones, and the engineering approach, which compares the canonical code with the best possible alternative. RESULTS: Here we used a genetic algorithm to search for better adapted hypothetical codes and as a method to guess the difficulty in finding such alternative codes, allowing to clearly situate the canonical code in the fitness landscape. This novel proposal of the use of evolutionary computing provides a new perspective in the open debate between the use of the statistical approach, which postulates that the genetic code conserves amino acid properties far better than expected from a random code, and the engineering approach, which tends to indicate that the canonical genetic code is still far from optimal. We used two models of hypothetical codes: one that reflects the known examples of codon reassignment and the model most used in the two approaches which reflects the current genetic code translation table. Although the standard code is far from a possible optimum considering both models, when the more realistic model of the codon reassignments was used, the evolutionary algorithm had more difficulty to overcome the efficiency of the canonical genetic code. CONCLUSIONS: Simulated evolution clearly reveals that the canonical genetic code is far from optimal regarding its optimization. Nevertheless, the efficiency of the canonical code increases when mistranslations are taken into account with the two models, as indicated by the fact that the best possible codes show the patterns of the standard genetic code. Our results are in accordance with the postulates of the engineering approach and indicate that the main arguments of the statistical approach are not enough to its assertion of the extreme efficiency of the canonical genetic code.

In this work we use the evolutionary multiobjective methodology for the optimization of topological active models, a deformable model that integrates features of region-based and boundary-based segmentation techniques. The model deformation is controlled by energy functions that must be minimized. As in other deformable models, a correct segmentation is achieved through the optimization of the model, governed by energy parameters that must be experimentally tuned. Evolutionary multiobjective optimization gives a solution to this problem by considering the optimization of several objectives in parallel. Concretely, we use the SPEA2 algorithm, adapted to our application, the search of the Pareto optimal individuals. The proposed method was tested on several representative images from different domains yielding highly accurate results.

In this article we describe the use of Artificial Neural Networks (ANN) with synaptic time delays between the nodes as a means to increase the capabilities of the usual control modules used in behavior based robotics. This inclusion allows the controllers to manage explicit temporal information in different levels. In the sensing level it permits the use of virtual sensors that improve the precision of the information provided by sensors through a temporal correlation of their values. In the actuation level we use the network with an infrasensorized robot in a problem that requires active sensing, where the control and actuation mechanisms are coordinated in order to obtain a better sensorial image of the environment by means of a spatio-temporal representation of a perception sequence. The decision of the appropriate delays is automated through learning and evolution.

In the context of visual systems for robots, we have made use of a high order gaussian synapses network and the Gaussian Synapses Backpropagation Algorithm (GSBP) for the implementation of the detectors that constitute one part of the whole visual architecture. These detectors are trained to be sensitive to spatial patterns that are relevant for the decisions the robot must perform during its operation in an environment. The inclusion of gaussian functions in the synapses of the network allows the network to select the appropriate spatial information and filter out all that is irrelevant according to the training it has received. In this paper we will show how these networks are easily trained to ignore backgrounds. In addition, with a very simple training set and an appropriate input selection strategy, the networks detect objects independently of size and position. These systems, coupled with an attention mechanism result in a very efficient visual information processor.

The aim of this paper is to endow a well-known structure for processing time-dependent information, synaptic delay-based ANN's, with a reliable and easy to implement algorithm suitable for training temporal decision processes. In fact, we extend the backpropagation algorithm to discrete-time feedforward networks that include adaptable internal time delays in the synapses. The structure of the network is similar to the one presented by [1], that is, in addition to the weights modeling the transmission capabilities of the synaptic connections, we model their length by means of a parameter that indicates the delay a discrete-event suffers when going from Zthe origin neuron to the target neuron through a synaptic connection. Like the weights, these delays are also trainable, and a training algorithm can be derived that is almost as simple as the backpropagation algorithm, and which is really an extension of it. We present examples of the application of these networks and algorithm to the prediction of time series and to the recognition of patterns in electrocardiographic signals. In the first case, we employ the temporal reasoning characteristics of these networks for the prediction of future values in a benchmark example of a time series: the one governed by the Mackey-Glass chaotic equation. In the second case, we provide a real life example. The problem consists in identifying different types of beats through two levels of temporal processing, one relating the morphological features which make up the beat in time and another one that relates the positions of beats in time, that is, considers rhythm characteristics of the ECG signal. In order to do this, the network receives the signal sequentially, no windowing, segmentation, or thresholding are applied.