Machine learning approaches reveal genomic regions associated with sugarcane brown rust resistance

Abstract

Sugarcane is an economically important crop, but its genomic complexity has hindered advances in molecular approaches for genetic breeding. New cultivars are released based on the identification of interesting traits, and for sugarcane, brown rust resistance is a desirable characteristic due to the large economic impact of the disease. Although marker-assisted selection for rust resistance has been successful, the genes involved are still unknown, and the associated regions vary among cultivars, thus restricting methodological generalization. We used genotyping by sequencing of full-sib progeny to relate genomic regions with brown rust phenotypes. We established a pipeline to identify reliable SNPs in complex polyploid data, which were used for phenotypic prediction via machine learning. We identified 14,540 SNPs, which led to a mean prediction accuracy of 50% when using different models. We also tested feature selection algorithms to increase predictive accuracy, resulting in a reduced dataset with more explanatory power for rust phenotypes. As a result of this approach, we achieved an accuracy of up to 95% with a dataset of 131 SNPs related to brown rust QTL regions and auxiliary genes. Therefore, our novel strategy has the potential to assist studies of the genomic organization of brown rust resistance in sugarcane.

Genome-wide approaches for the identification of markers and genes associated with sugarcane yellow leaf virus resistance

“…We could achieve very high accuracies of prediction (up to 95%) with considerably reduced datasets comprising 120-190 markers. These results are very similar to what was obtained for predicting sugarcane brown rust resistance groups, where an accuracy of 95% was obtained using 131 SNPs 62 . Marker datasets selected by ML have rarely been employed in genetic association studies in plants, but the few existing examples show their power to identify genes associated with phenotypes of interest [80][81][82] .…”

“…For genome-wide mapping of quantitative traits – for which a high density of markers is necessary – codominant SNPs and indels are currently more cost-effective, as they can be easily identified in much larger numbers. Additionally, in the case of polyploids, these markers offer the possibility of estimating allele dosages or APs, which are highly informative for genetic studies (de Bem Oliveira, 2019; Matias et al, 2019; Aono et al, 2020). Hence, we believe the results we obtained with codominant SNPs and indels are more reliable, as they rely on much larger amounts of genetic information.…”

“…In this crop, the most relatable findings available are those reported by Creste et al 59 , who showed that using different dominant markers can bias genetic analyses, and thus the choice of marker must be guided by the specific goal of each study. For GWASsfor which a high density of markers is usually necessary -SNPs and indels are currently more cost-effective, as they can be easily identified in much larger numbers, in addition to offering the possibility of estimating highly-informative allele dosages or APs [60][61][62] . Hence, we believe the results we obtained with codominant SNPs and indels are more reliable, as they lean on much more genetic information.…”

“…Finally, we assessed the capacity of ML strategies to predict the attribution of genotypes to the phenotypic groups identified in the HCPC analyses based on all markers, following the genomic prediction approach proposed by Aono et al (2020). For this approach, we selected accessions successfully genotyped with both SNPs/indels and AFLPs/SSRs; missing data in dominant markers were imputed as the means.…”

“…Markers previously associated with SCYLV resistance by QTL mapping 26,29 and GWAS 27,30 were also retrieved and included in the map. Finally, the sequences of associated markers were annotated by aligning SSR flanking sequences or the 2000-bp window adjacent to SNPs and indels against a database comprising CDSs of the genomes of 14 Poaceae species and Arabidopsis thaliana 62 . For this, BLASTn 183 was used with an E-value of 1e-30, and the best alignment of each sequence was kept for analysis.…”