Software

Explores trial and germplasm data. For a trial with one row per plot: analysis of variance for one or two treatment factors, completely randomized or in complete blocks, with Tukey HSD, Fisher LSD or Duncan mean comparison letters, Shapiro-Wilk and Levene assumption checks, means plots, and a summary table across several traits at once giving each treatment mean with its letter, the LSD and the CV. For a genotype × trait table: data overview and validation, correlation analysis between traits, PCA and hierarchical clustering (via FactoMineR/factoextra), and multi-trait genotype selection with the MGIDI index (via the metan package).

Reference R implementation of the Stage-3 pilot genomic-selection workflow in Table 5 of the Ecological Genetics and Genomics review on AI-assisted breeding in sweetpotato. The pipeline calls dosage-uncertain genotypes with polyRAD, builds a hexaploid additive relationship matrix with AGHmatrix (VanRaden extended to ploidy 6), fits dosage-aware GBLUP for five target traits under five-fold cross-validation with sommer, computes a classical MGIDI Stage-1 baseline on the same traits with metan, and tests the accuracy difference between them with a 10,000-resample bootstrap confidence interval. No genotyping-by-sequencing data yet exist for the sweetpotato panel, so all genotypes and phenotypes in the repository are simulated (a 315-progeny biparental hexaploid population); only the data-loading step changes when real genotyping data become available.

Companion app for SCIFI (Sweet Corn Ideotype Fitness Index), which consolidates 21 independently run sweet- and specialty-corn breeding trials (927 genotype-records, Vietnam National University of Agriculture, 2018–2024) that share almost no common germplasm, trait list, or selection method into one comparable score. Ten agronomic and quality criteria (yield, Brix, pericarp thickness, ear length and diameter, kernel rows per ear, kernels per row, plant height, ear height, and days to fresh harvest) are each converted to a desirability score and combined by a non-compensatory geometric mean, with boundaries calibrated from pooled cross-trial percentiles. Parental inbred lines and their own F1 hybrids score systematically differently on heterosis-sensitive traits, so the app offers three class-calibrated scoring scales: pooled, hybrid, and parental line. It also compares SCIFI with the selection methods the program used before (MGIDI, MTSI, WAASB, Smith-Hazel index, Smith index), and runs descriptive statistics, ANOVA, and PCA on a new trial’s raw data before scoring it. Includes an interactive calculator and a searchable table of the 871 genotype-by-trial SCIFI scores. Manuscript in preparation.