Seleção e avaliação de marcadores moleculares com grande informatividade para a predição do valor genômico

Loading...
Thumbnail Image

Journal Title

Journal ISSN

Volume Title

Publisher

Universidade Federal de Juiz de Fora (UFJF)

DOI

Abstract

The animal breeding seeks to maximize of a characteristic over time. The use of genomic data gave rise that we know as genomic selection. The made of efficient models for evaluate the merit of an animal is complex and generally is based on the assumption of additive genetic effects. However, in the presence of non-additive genetic variability, the models may not contain all the complexity of possible interactions between the genes, the epistasis. The S4GS is a genomic data simulator developed in this thesis, which seeks to mimic important features for the study in genomic selection such as linkage disequilibrium, artificial insemination and crossing over. Another relevant factor is the ability to simulate different gene actions and interactions at multiple levels. It was used in the creation of 8 study scenarios, highlighting the scenario 8 that sought to simulate the Girolando option B. The proposed method consists of a two-step selection and evaluation, generating an optimal combination for the increase of accuracy. The algorithms chosen for the feature selection step were: the FFS that was developed in this thesis; The SMS; and CART as a quick alternative. The evaluation stage used two classical techniques, the RR-BLUP and the BLASSO as a reference, and the SVR. The association of the techniques used in the selection and evaluation stage leads us to three models: SVR + FFS; SVR + SMS; and SVR + CART. In the results obtained, the selection of attributes proved to be an important resource in increasing accuracy in all 8 scenarios. The simulation process allowed data to be obtained up to 15th generation allowing the models generated in 1th or 4th to in subsequent ones to be applied. The application of feature selection significantly increased accuracy in genomic data, except for the small sample set and in completely linear data. The proposed method was able to the bases with the descriptive characteristics to be efficient, generating a significant increase in the final correlation.

Description

Citation

Endorsement

Review

Supplemented By

Referenced By

Rights and licensing

Acesso Aberto