Publications des agents du Cirad


The Genomic Selection of Theobroma cacao: a new strategy of marker assisted selection to improve breeding efficiency and predict useful traits in new populations. [20]

Ribeyre F., Sounigo O., Argout X., Cilas C., Efombagn M.I.B., Denis M., Bouvet J.M., Fouet O., Lanaud C.. 2017. In : Proceedings of the International Symposium on cocoa research. Lima : ICCO, 7 p.. International Symposium on Cocoa Research (ISCR 2017). 1, 2017-11-13/2017-11-17, Lima (Pérou).

Genomic selection (GS) is a statistical approach that utilizes all available genome-wide markers simultaneously and phenotypic traits of a ¿training population¿ to estimate breeding values or total genetic values. For breeding programs, GS is a promising alternative to the traditional marker-assisted selection for manipulating complex polygenic traits often controlled by many small effect genes. A major interest of this method is also to be able to make predictions of trait values, from a training population, on trees only genotyped by molecular markers The use of the appropriate statistical model remains one of the critical issues of the GS. The relative performance of biometrical models is expected to depend on the genetic background of the traits under assessment. The objective of this study was to estimate the reliability of different models of genomic selection to predict two agronomic traits of cacao - yield and resistance to Phytophthora pod rot. The study was performed on 287 trees present in a cacao farm plot in Cameroon, belonging to several full-sib progenies released to farmers as commercial varieties. Each tree was genotyped, using more than 5,000 GBS (genotyping by sequencing) based on SNP markers, and assessed for the mean average of one bean, a trait contributing to cocoa yield, and the % of rotten, as a measure of resistance to Phytophthora megakarya. Two models were used: Best linear unbiased prediction model and Bayesian lasso model. Cross-validation was used to test their predictive ability. It is an assumption-free method using an estimation set for model training and an independent test set for prediction. Predictive ability of models was good for both traits indicating that GS is a promising method to improve these cocoa traits. However, it was slightly higher for average weight of a bean (R= 0.59) than for % of rotten pods (R= 0.42).

Documents associés

Communication de congrès

Agents Cirad, auteurs de cette publication :