Carlos Cristiano Hasenclever Borges

SNPs selection using support vector regression and genetic algorithms in GWAS

IntroductionThis paper proposes a new methodology to simultaneously select the most relevant SNPs markers for the characterization of any measurable phenotype described by a continuous variable using Support Vector Regression with Pearson Universal kernel as fitness function of a binary genetic algorithm. The proposed methodology is multi-attribute towards considering several markers simultaneously to explain the phenotype and is based jointly on statistical tools, machine learning and computational intelligence.ResultsThe suggested method has shown potential in the simulated database 1, with additive effects only, and real database. In this simulated database, with a total of 1,000 markers, and 7 with major effect on the phenotype and the other 993 SNPs representing the noise, the method identified 21 markers. Of this total, 5 are relevant SNPs between the 7 but 16 are false positives. In real database, initially with 50,752 SNPs, we have reduced to 3,073 markers, increasing the accuracy of the model. In the simulated database 2, with additive effects and interactions (epistasis), the proposed method matched to the methodology most commonly used in GWAS.ConclusionsThe method suggested in this paper demonstrates the effectiveness in explaining the real phenotype (PTA for milk), because with the application of the wrapper based on genetic algorithm and Support Vector Regression with Pearson Universal, many redundant markers were eliminated, increasing the prediction and accuracy of the model on the real database without quality control filters. The PUK demonstrated that it can replicate the performance of linear and RBF kernels.

A Semi-deterministic Ensemble Strategy ForImbalanced Datasets (SDEID) Applied ToBankruptcy Prediction

In the last decade, there was a rapid growth in the availability and use of credit for Brazilian companies. Until recently, the decision to grant credit was based on human trial to evaluate the risk of insolvency. Increased demand from companies for credit has led to the use of more accurate models for bankruptcy prediction. In recent years much progress has occurred in the process of drawing up a model fostered by increased competition among financial institutions, changes in the economic environment for businesses and advances in computational techniques. This article discusses and presents alternatives for some of the main problems in the preparation of models for bankruptcy prediction with the application of data mining techniques. The first problem approached is the class imbalance that may cause a poor classification performance and it is treated jointly with an ensemble strategy. The other one rely on the selection of the most significant combination of attributes, the financial variables, which have been widely studied in insolvency prediction. Finally, it is presented a case study in a real world data base of Brazilian companies.

Perceptron models for online structured prediction

Our structured prediction problem is formulated as a convex optimization problem of maximal margin [5-6], quite similar to the formulation of multiclass support vector machines (MSVM) [8]. It is applied to predict costs among states of paths. Predicting them properly is very important, because the problem of paths planning depends on its correctness. Ratliff [4] showed a maximum margin approach which allows the prediction of costs in different environments using subgradient method. As a contribution of this work, we developed new solution methods: the first one, called Structured Perceptron, has similarities with the correction scheme proposed by [1] and the second one is called Structured IMA. It is derived from the work presented by [2]. Both use the Perceptron model. The proposed algorithms were more efficient in terms of computational effort and similar in prediction quality when compared with [4].

An initialization method for grammatical evolution assisted by decision trees

Grammatical Evolution (GE) is a genetic programming technique in which the candidate solutions are represented using a binary genotype and the programs can be generated through production rules of a formal grammar. Similarly to other evolutionary computation methods, the GEs performance can be improved when an adequate initial population seeding is adopted. Decision trees are widely used to model classifiers in machine learning and their symbolic form can be mapped back to the GEs binary representation of the candidate individuals. Thus, the use of machine learning techniques to generate decision trees to compose the initial population of GE is investigated here. Computational experiments with a real world data set are carried out and the results show an increase of performance when compared to the traditional seeding approach.

Structural Comparative Analysis of Ecto- NTPDase Models from S. Mansoni and H. Sapiens

The control of extracellular nucleoside concentrations by Extracellular Nucleoside Triphosphate Diphosphohydrolases (Ecto- NTPDases) is essential in the regulation of the purinergic signalling and also in immune response. In mammals, four isoforms of Ecto-NTPDases have been described (NTPDase1-3 and NTPDase8). The isoform 1 of human Ecto-NTPDase (HsNTPDase1 or CD39) is expressed in endothelial cells of veins and arteries. An Ecto-NTPDase have been identified in the tegument of adult worms of Schistosoma mansoni (SmNTPDase1), and it was located on the outer surface of parasite’s tegument. Due to the location of the SmATPDase1, it was proposed that these Ecto-NTPDase participate in the evasion of the host immune system by parasite. These assumptions reinforce the importance of researching the SmATPDase1 as a drug target candidate for the schistosomiasis treatment. In this work, we propose the three-dimensional structure model of the enzymes SmATPDase1 and CD39 using comparative modeling. The results show similarities between these proteins, especially in the active site region, and become necessary to search for alternative binding site of drugs aiming new therapies for schistosomiasis.

A Genetic Programming Model for Association Studies to Detect Epistasis in Low Heritability Data

The genome-wide associations studies (GWAS) aims to identify the most influential markers in relation to the phenotype values. One of the substantial challenges is to find a non-linear mapping between genotype and phenotype, also known as epistasis, that usually becomes the process of searching and identifying functional SNPs more complex. Some diseases such as cervical cancer, leukemia and type 2 diabetes have low heritability. The heritability of the sample is directly related to the explanation defined by the genotype, so the lower the heritability the greater the influence of the environmental factors and the less the genotypic explanation. In this work, an algorithm capable of identifying epistatic associations at different levels of heritability is proposed. The developing model is a aplication of genetic programming with a specialized initialization for the initial population consisting of a random forest strategy. The initialization process aims to rank the most important SNPs increasing the probability of their insertion in the initial population of the genetic programming model. The expected behavior of the presented model for the obtainment of the causal markers intends to be robust in relation to the heritability level. The simulated experiments are case-control type with heritability level of 0.4, 0.3, 0.2 and 0.1 considering scenarios with 100 and 1000 markers. Our approach was compared with the GPAS software and a genetic programming algorithm without the initialization step. The results show that the use of an efficient population initialization method based on ranking strategy is very promising compared to other models.

A dual method for solving the nonlinear structured prediction problem

A primal structured perceptron method to solve the linear structured prediction problem.A kernel method to solve the nonlinear structured prediction problem based on sampling.An alternative approach to solve the inverse reinforcement problem.An efficient solution for the path planning prediction costs problem. In this paper, we present a perceptron-based algorithm and have developed a dual formulation to solve the nonlinear structured prediction problem, which we called Dual Structured Incremental Margin Algorithm (DSIMA). The proposed formulation allows the introduction of kernel functions enabling the efficient solution of nonlinear problems. In order to verify the correctness and applicability of the algorithm, we consider an inverse approach to the path planning problem. The problem mapped on a grid environment can be solved by a search process that essentially depends on the definition of the transition costs between states. In this context, we develop and apply a learning algorithm that is able to perform the reverse path, i.e., the prediction of these costs in a direct space for the linear form. However, considering the nonlinear form, the problem is solved in a space of high dimension and where it is possible to learn a path instead of the transition costs. This learning problem is usually formulated as a convex optimization problem of maximum margin. Several tests to solve the costs prediction problem were carried out and the results compared to other structured prediction techniques. The proposed algorithm demonstrated greater efficiency in terms of computational effort and quality of prediction.

Statistical signal similarity check using symbolic data for power management on low capacity devices

Considering the success of mobile computing, realtime identification of Passive Acoustic Monitoring (PAM) data is now an emerging possibility. Despite computational evolution, analysis of raw acoustic data by complex algorithms requires considerably computing effort, therefore, consuming overly battery power. As battery power is a low resource in many environments, such as the sea, a very simple time-domain signal similarity filter is proposed in this paper. To accomplish that, the filter uses a richer representation of time-domain data created by symbolic data analysis. Taking this new data type and assuming environmental noise as a stationary process, a non-parametric statistical hypothesis test is applied to detect signal similarity over time. To evaluate overall processing time, a dataset of raw acoustic data acquired from PAM was used. In addition, to endorse accuracy and no data loss, all data were visually and acoustically searched for sperm whale clicks.

On the robustness of SNPs filtering using computational intelligence

This work uses a filter based on neural networks to verify the mismatches in two Arabidopsis thaliana germplasm. Aiming to demonstrate the robustness and adaptability of the filter it will be applied in a reuse model context. The neural network filter previously defined and performed using the genome of an animal of the species Bos Taurus is used maintaining the main parameterization pre-defined to identify the SNPs on the mismatches detected in the reassembled germplasm. The experiments with the adapted filter in the new genome indicate that the quality and level of SNPs detection are preserved despite of the lack of a training process for this specific data.

Decision support in attribute selection with machine learning approach

This paper proposes a method to simultaneously select the most relevant single nucleotide polymorphisms (SNPs) markers - the attributes - for the characterization of any measurable phenotype described by a continuous variable using support vector regression (SVR) with Pearson VII Universal Kernel (PUK). The proposed study is multiattribute towards considering several markers simultaneously to explain the phenotype and is based jointly on a statistical tools, machine learning and computational intelligence.