Search

Graphical Abstract

The objective of the present study was to evaluate the accuracy and bias of direct and blended genomic predictions using different methods and cross-validation techniques for growth traits (weight and weight gains) and visual scores (conformation, precocity, muscling, and size) obtained at weaning and at yearling in Hereford and Braford breeds. Phenotypic data contained 126,290 animals belonging to the Delta G Connection genetic improvement program, and a set of 3,545 animals genotyped with the 50K chip and 131 sires with the 777K. After quality control, 41,045 markers remained for all animals. An animal model was used to estimate (co)variance components and to predict breeding values, which were later used to calculate the deregressed estimated breeding values (DEBV). Animals with genotype and phenotype for the traits studied were divided into 4 or 5 groups by random and k-means clustering cross-validation strategies. The values of accuracy of the direct genomic values (DGV) were moderate to high magnitude for at weaning and at yearling traits, ranging from 0.19 to 0.45 for the k-means and 0.23 to 0.78 for random clustering among all traits. The greatest gain in relation to the pedigree BLUP (PBLUP) was 9.5% with the BayesB method with both the k-means and the random clustering. Blended genomic value accuracies ranged from 0.19 to 0.56 for k-means and from 0.21 to 0.82 for random clustering. The analyses using the historical pedigree and phenotypes contributed additional information to calculate the GEBV, and in general, the largest gains were for the single-step (ssGBLUP) method in bivariate analyses with a mean increase of 43.00% among all traits measured at weaning and of 46.27% for those evaluated at yearling. The accuracy values for the marker effects estimation methods were lower for k-means clustering, indicating that the training set relationship to the selection candidates is a major factor affecting accuracy of genomic predictions. The gains in accuracy obtained with genomic blending methods, mainly ssGBLUP in bivariate analyses, indicate that genomic predictions should be used as a tool to improve genetic gains in relation to the traditional PBLUP selection.

In this paper, we introduce featsel, a framework for benchmarking of feature selection algorithms and cost functions. This framework allows the user to deal with the search space as a Boolean lattice and has its core coded in C++ for computational efficiency purposes. Moreover, featsel includes Perl scripts to add new algorithms and/or cost functions, generate random instances, plot graphs and organize results into tables. Besides, this framework already comes with dozens of algorithms and cost functions for benchmarking experiments. We also provide illustrative examples, in which featsel outperforms the popular Weka workbench in feature selection procedures on data sets from the UCI Machine Learning Repository.

Highlights • The U-curve problem can be used to model feature selection. • We pointed out an error in the first algorithm proposed to solve the U-curve problem. • We introduced UCS, an optimal algorithm for the U-curve problem. • We also provide UCSR, an algorithm for a special case of the U-curve problem. • UCS and UCSR outperformed BFS, CHCGA and SFFS in feature selection experiments. Abstract The U-curve optimization problem is characterized by a decomposable in U-shaped curves cost function over the chains of a Boolean lattice. This problem can be applied to model the classical feature selection problem in Machine Learning. In this paper, we point out that the firstly proposed algorithm to tackle the U-curve problem, the RBM algorithm, is in fact suboptimal. We also present two new algorithms: UCS, which is actually optimal to tackle this problem; and UCSR, a variation of UCS that solves a special case of the U-curve problem and relies on a reduced, ordered binary decision diagram to control the search space. We provide results of two computational assays with these new algorithms: first, W-operator design for filtering of binary images; second, linear SVM design for classification of data sets from the UCI Machine Learning Repository. We show that, in these assays, UCS and UCSR outperformed an exhaustive search and also three widely used heuristics: the SFFS sequential selection, the BFS graph-based search, and the CHCGA genetic algorithm. Finally, we analyze the obtained results and point out improvements that might enhance the performance of these two novel algorithms. [ABSTRACT FROM AUTHOR]