PT  - JOURNAL ARTICLE
AU  - Marnin D. Wolfe
AU  - Dunia Pino Del Carpio
AU  - Olumide Alabi
AU  - Chiedozie Egesi
AU  - Lydia C. Ezenwaka
AU  - Ugochukwu N. Ikeogu
AU  - Robert S. Kawuki
AU  - Ismail S. Kayondo
AU  - Peter Kulakow
AU  - Roberto Lozano
AU  - Ismail Y. Rabbi
AU  - Esuma Williams
AU  - Alfred A. Ozimati
AU  - Jean-Luc Jannink
TI  - Prospects for genomic selection in cassava breeding
AID  - 10.1101/108662
DP  - 2017 Jan 01
TA  - bioRxiv
PG  - 108662
4099  - http://biorxiv.org/content/early/2017/02/14/108662.short
4100  - http://biorxiv.org/content/early/2017/02/14/108662.full
AB  - Cassava (Manihot esculenta Crantz) is a clonally propagated staple food crop in the tropics. Genomic selection (GS) reduces selection cycle times by the prediction of breeding value for selection of unevaluated lines based on genome-wide marker data. GS has been implemented at three breeding programs in sub-Saharan Africa. Initial studies provided promising estimates of predictive abilities in single populations using standard prediction models and scenarios. In the present study we expand on previous analyses by assessing the accuracy of seven prediction models for seven traits in three prediction scenarios: (1) cross-validation within each population, (2) cross-population prediction and (3) cross-generation prediction. We also evaluated the impact of increasing training population size by phenotyping progenies selected either at random or using a genetic algorithm. Cross-validation results were mostly consistent across breeding programs, with non-additive models like RKHS predicting an average of 10% more accurately. Accuracy was generally associated with heritability. Cross-population prediction accuracy was generally low (mean 0.18 across traits and models) but prediction of cassava mosaic disease severity increased up to 57% in one Nigerian population, when combining data from another related population. Accuracy across-generation was poorer than within (cross-validation) as expected, but indicated that accuracy should be sufficient for rapid-cycling GS on several traits. Selection of prediction model made some difference across generations, but increasing training population (TP) size was more important. In some cases, using a genetic algorithm, selecting one third of progeny could achieve accuracy equivalent to phenotyping all progeny. Based on the datasets analyzed in this study, it was apparent that the size of a training population (TP) has a significant impact on prediction accuracy for most traits. We are still in the early stages of GS in this crop, but results are promising, at least for some traits. The TPs need to continue to grow and quality phenotyping is more critical than ever. General guidelines for successful GS are emerging. Phenotyping can be done on fewer individuals, cleverly selected, making for trials that are more focused on the quality of the data collected.(GS)Genomic selection(GBS)genotype-by-sequencing(IITA)International Institute of Tropical Agriculture(NRCRI)National Root Crops Research Institute(NaCRRI)National Crops Resources Research Institute(GEBVs)genomic estimated breeding values(TP)training population(RTWT)fresh root weight(RTNO)root number(SHTWT)fresh shoot weight(HI)harvest index(DM)dry matter(CMD)content cassava mosaic disease(MCMDS)mean CMD severity(VIGOR)early vigor