Enhancing Selection Intensity with High-Throughput Phenotyping Data in Plant Breeding

Abstract

Genomic prediction (GP) has proven to be an essential tool to accelerate the cultivar development pipeline by predicting the performance of un-phenotyped lines in the field of plant breeding. However, the prediction accuracy is bounded by the heritability of the target trait, limiting the efficiency of the single-trait GP model. To overcome the limitation, a multi-trait genomic prediction model using high-throughput phenotyping (HTP) data can be used to leverage secondary traits to boost the predictive ability of the target trait. This study aimed to assess the efficiency of the multi-trait (MT)- GP model powered by HTP derived secondary traits in predicting nitrogen leaf area (Narea), specific leaf area (SLA), partial least square regression (PLSR)-Narea, and PLSR-SLA in sorghum. Three secondary traits (S1, S2, S3) were identified by using whole spectra of hyperspectral data using co heritability measures and were named synthetic traits. As a baseline model, single-trait GBLUP(GenomicBestLinearUnbiasedPredictor) was fitted, followed by three MT-GBLUP models using synthetic traits and target traits together. Model performances were assessed using k-fold (k=5) cross-validation (CV) schemes which consisted of single-trait, CV1, and CV2 schemes. The heritability of Narea, SLA, PLSR-Narea, and PLSR-SLA was 0.32,0.34, 0.40, and 0.26, respectively. Additionally, the heritability ranged from 0.61– 0.68, the genetic correlation ranged from 0.7– 0.9, and the co-heritability ranged from 0.46-0.57 across synthetic traits selected for four target traits. The high genetic correlation and heritability of synthetic traits met the requirements for their use as a secondary trait. The use of synthetic traits in the MT-GP model enhanced the accuracy of prediction by 6 %, 10 %, 10.87%, and 7.5% compared to a single trait alone in Narea, SLA, PLSR-Narea, PLSR-SLA respectively. Furthermore, the study demonstrated an improvement in prediction accuracy while using secondary traits derived from HTP in the MT-GP model compared to a single trait using Narea and SLA alone. Overall, our analysis highlights a practical approach to leverage high throughput phenotyping data to improve the performance of genomic prediction models.

Plant phenotyping relevance

高スループット・ハイパースペクトルデータから合成形質を抽出し、交差検証でゲノム予測への有効性を評価しており、表現型取得・推定ワークフローが中心である。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.