Integrating whole-genome resequencing and machine learning to refine QTL analysis for fruit quality traits in peach

Increasing marker density results in better map coverage and efficiency of genetic analysis. Here, we resequenced a large (<i>N</i> = 235) F1 progeny from two distant peach cultivars, 'Zhongyou Pan #9' and 'September Free', and constructed two parental maps (1:1 segregations) and one combined map (1:2:1 segregations) with 134 277 SNPs. Markers with the same genotype for all individuals studied were grouped in bins and a unique genotype for each bin was inferred to avoid mapping problems derived from erroneous data. The total genetic distance of the two parental maps was 431.9 and 594.2 cM with a short mean distance, 0.9 cM, between contiguous bins (groups of markers with the same genotype) and high collinearity with the peach genome. The genetics of eight fruit-related traits was analyzed for 2 years, allowing the positions of two major genes, fruit shape (<i>S</i>) and flesh adhesion to the stone (<i>F</i>), to be established, along with nine quantitative trait loci (QTLs) for quantitative traits including fruit soluble solids concentration, titratable acidity, weight, maturity date, and flesh color (yellow to orange). We developed a machine learning-based linear model to assess flesh color, which proved more efficient than physical colorimetric parameters (L, a^*, b^*), detecting consistent QTLs. Based on map position, gene expression patterns, and function, candidate genes were identified. Overall, our results provide two new elements: ultra-high-density maps with resequencing data to enhance mapping resolution and phenotyping strategies based on machine learning models that improve the quality of quantitative measurements to help understand the genetic control of key fruit quality traits.