Evaluating dimensionality reduction for genomic prediction

Abstract

The development of genomic selection (GS) methods has allowed plant breeding programs to select favorable lines using genomic data before performing field trials. Improvements in genotyping technology have yielded high-dimensional genomic marker data which can be difficult to incorporate into statistical models. In this paper, we investigated the utility of applying dimensionality reduction (DR) methods as a pre-processing step for GS methods. We compared five DR methods and studied the trend in the prediction accuracies of each method as a function of the number of features retained. The effect of DR methods was studied using three models that involved the main effects of line, environment, marker, and the genotype by environment interactions. The methods were applied on a real data set containing 315 lines phenotyped in nine environments with 26,817 markers each. Regardless of the DR method and prediction model used, only a fraction of features was sufficient to achieve maximum correlation. Our results underline the usefulness of DR methods as a key pre-processing step in GS models to improve computational efficiency in the face of ever-increasing size of genomic data.

A tied-weight autoencoder for the linear dimensionality reduction of sample data

“…Dimensionality reduction finds applications in various domains such as feature selection, computational efficiency, efficient data storage, and noise reduction in data visualization. Moreover, it is instrumental in addressing statistical modeling and inference challenges that arise in cases characterized by the "large p, small n" problem, where the sample dimensions p significantly exceeds the sample size n. For example, dimensionality reduction becomes necessary in cases such as microarray data analysis 3 and genomic selection 4 , where there are a large number of genetic variants (typically in the tens of thousands), but the number of samples available for each gene is relatively limited. As a crucial aspect of data preprocessing, dimensionality reduction methods are of great importance in the fields of machine learning and data science 5,6 .…”

Knowledge-driven approaches to improve genomic prediction in plants

“…Despite significant progress in high-throughput phenotyping (HTP) during the last decade, phenotyping plants in comparable numbers to those of the SNPs is still not practical. This results in a situation often termed "the curse of dimensionality" (Manthena, Jarquín et al 2022), where the number of observations n is far smaller than the number of variables p (n << p). As a consequence, the models become very complex with millions of estimable parameters, are prone to overfit on the limited training data and do not generalise well on the test data.…”

“…An approach to remove highly correlated SNPs uses binning of contiguous regions on the genome and selects only one SNP out of a bin (Du, Wei et al 2018), but an optimal selection of bin size is not straightforward. Dimensionality reduction can also be achieved by transformation of the genotype matrix into a low-dimensional representation (Manthena, Jarquín et al 2022). For instance, singular value decomposition of the genotype matrix can be employed to perform GP using a smaller 1 number of principal components as features instead of using all SNPs (Pintus, Gaspa et al 2012, Du, Wei et al 2018, Odegard, Indahl et al 2018.…”

“…Consequently, input pruning is critically important, especially for complex models like (deep) neural networks (Abdollahi-Arpanahi, Gianola et al 2020). Among different approaches proposed for dimensionality reduction (Manthena, Jarquín et al 2022), we used variable importance scores obtained from random forests, where a positive value implies association to the phenotype (Li, Raidan et al 2019). We compared calculating these on all SNPs (Wang, Aggarwal et al 2017) (RF-PRIORNET, RF-MLP) or for each SNP group (GROUPRF-PRIORNET).…”