Minor QTLs mining through the combination of GWAS and machine learning feature selection

Abstract

2 Introduction: Minor QTLs mining has a very important role in genomic selection, pathway analysis and 3 trait development in agricultural and biological research. Since most individual loci contribute little to 4 complex trait variations, it remains a challenge for traditional statistical methods to identify minor QTLs 5 with subtle phenotypic effects. Here we applied a new framework which combined the GWAS analysis 6 and machine learning feature selection to explore new ways for the study of minor QTLs mining. 7 Results: We studied the soybean branching trait with the 2,137 accessions from soybean (Glycine max) 8 diversity panel, which was sequenced by 50k SNP chips with 42,080 valid SNPs. First as a baseline 9 study, we conducted the GWAS GAPIT analysis, and we found that only one SNP marker significantly 10 associated with soybean branching was identified. We then combined the GWAS analysis and feature 11 importance analysis with Random Forest score analysis and permutation analysis. Our analysis results 12 showed that there are 36,077 features (SNPs) identified by Random Forest score analysis, and 2,098 13 features (SNPs) identified by permutation analysis. In total, there are 1,770 features (SNPs) confirmed by 14 both of the Random Forest score analysis and the permutation analysis. Based on our analysis, 328 15 branching development related genes were identified. A further analysis on GO (gene ontology) term16 enrichment were applied on these 328 genes. And the gene location and gene expression of these 17 identified genes were provided. 18Conclusions: We find that the combined analysis with GWAS and machine learning feature selection 19 shows significant identification power for minor QTLs mining. The presented research results on minor 20 QTLs mining will help understand the biological activities that lie between genotype and phenotype in 21 terms of causal networks of interacting genes. This study will potentially contribute to effective genomic 22 selection in plant breeding and help broaden the way of molecular breeding in plants. 23

Multiomic investigation of sugarcane mosaic virus resistance in sugarcane

“…Thus, it is possible that this marker is linked to Sspon.02G0027920-1A, the S. spontaneum gene syntenic to the auxin-binding protein gene at Scmv2 (Ding et al, 2012). This is an indication of the potential suitability of FS methodologies for the identification of QTLs, which is supported by other studies in which researchers analyzed traits controlled by many loci (Zhou et al, 2019).…”

Identification of Age-Specific and Common Key Regulatory Mechanisms Governing Eggshell Strength in Chicken Using Random Forests

“…Unlike these traditional statistical models, machine learning methods do not require these prior assumptions about the genetic architecture of traits and have been applied in GWAS in humans [30] as well as in livestock [27,95]. Especially, Romagnoni et al [30] and Huang et al [24] showed that machine learning based algorithms provide promising prediction power to assess genotype-phenotype associations. In particular, the Random Forests (RF) algorithm has been successfully applied for this purpose.…”

“…To overcome these limitations of GWAS, application of Bayesian frameworks as well as machine learning algorithms have gained importance in the last decade [21][22][23][24][25]. Their comparative performance has been evaluated for a variety of traits with different genetic architectures (see the reviews [13,26,27]).…”

Exploring Machine Learning Algorithms to Unveil Genomic Regions Associated With Resistance to Southern Root-Knot Nematode in Soybeans

“…To select SNPs that were significantly associated with SRKN resistance, a Partial Least Square (PLS) ( Wold, 1966 ) model was fitted using the 4,974 SNPs as predictors and the number of galls in the root as responses. PLS models have the advantage to reduce the variability and instability of estimated responses caused by multicollinearity among predictors ( Zhou et al, 2019 ; James et al, 2021 ). Additionally, PLS creates linear combinations (known as components) of the original predictor variables (the SNPs) to explain the observed variability in the responses (the galling response).…”

“…Besides, the applications of ML-based GWAS need to be consistently validated with significant associations that make both biological and statistical sense ( Nicholls et al, 2020 ). To the best of our knowledge, ML-based GWAS has been applied in soybean to identify significant marker-trait associations using SVM ( Yoosefzadeh-Najafabadi et al, 2021a , b ), RF ( Zhou et al, 2019 ; Xavier and Rainey, 2020 ; Yoosefzadeh-Najafabadi et al, 2021b ), and Deep Convolutional Neural Network (CNN) ( Liu et al, 2019 ), of which none was applied on soybean resistance to SRKN. Therefore, the objective of this study was to conduct ML-GWAS utilizing 717 diverse breeding lines derived from 330 unique bi-parental populations with two different algorithms (SVM and RF) to unveil novel regions of the soybean genome regulating the resistance to SRKN (reported as the development of galls in the roots) and contribute to developing enhanced and more durable SRKN resistance.…”