MGIDI: toward an effective multivariate selection in biological experiments

Abstract

Motivation

Multivariate data are common in biological experiments and using the information on multiple traits is crucial to make better decisions for treatment recommendations or genotype selection. However, identifying genotypes/treatments that combine high performance across many traits has been a challenger task. Classical linear multi-trait selection indexes are available, but the presence of multicollinearity and the arbitrary choosing of weighting coefficients may erode the genetic gains.

Results

We propose a novel approach for genotype selection and treatment recommendation based on multiple traits that overcome the fragility of classical linear indexes. Here, we use the distance between the genotypes/treatment with an ideotype defined a priori as a multi-trait genotype–ideotype distance index (MGIDI) to provide a selection process that is unique, easy-to-interpret, free from weighting coefficients and multicollinearity issues. The performance of the MGIDI index is assessed through a Monte Carlo simulation study where the percentage of success in selecting traits with desired gains is compared with classical and modern indexes under different scenarios. Two real plant datasets are used to illustrate the application of the index from breeders and agronomists’ points of view. Our experimental results indicate that MGIDI can effectively select superior treatments/genotypes based on multi-trait data, outperforming state-of-the-art methods, and helping practitioners to make better strategic decisions toward an effective multivariate selection in biological experiments.

Availability and implementation

The source code is available in the R package metan (https://github.com/TiagoOlivoto/metan) under the function mgidi().

Supplementary information

Supplementary data are available at Bioinformatics online.