SpykProps: an imaging pipeline to quantify architecture in unilateral grass inflorescences.

Abstract

Background Inflorescence properties such length, spikelet number, and their spatial distribution across the rachis, are fundamental indicators of seed productivity in grasses and have been a target of selection throughout domestication and crop improvement. However, quantifying such complex morphology is laborious, time-consuming, and commonly limited to human-perceived traits. These limitations can be exacerbated by unfavorable trait correlations between inflorescence architecture and seed yield that can be unconsciously selected for. Computer vision offers an alternative to conventional phenotyping, enabling higher throughput and reducing subjectivity. These approaches provide valuable insights into the determinants of seed yield, and thus, aid breeding decisions. Results Here, we described SpykProps, an inexpensive Python-based imaging system to quantify morphological properties in unilateral inflorescences, that was developed and tested on images of perennial grass (Lolium perenne L.) spikes. SpykProps is able to rapidly and accurately identify spikes (RMSE 2 = 0.96), and number of spikelets (R 2 = 0.61). It also quantifies color and shape from hundreds of interacting descriptors that are accurate predictors of architectural and agronomic traits such as seed yield potential (R 2 = 0.94), rachis weight (R 2 = 0.83), and seed shattering (R 2 = 0.85). Conclusions SpykProps is an open-source platform to characterize inflorescence architecture in a wide range of grasses. This imaging tool generates conventional and latent traits that can be used to better characterize developmental and agronomic traits associated with inflorescence architecture, and has applications in fields that include breeding, physiology, evolution, and development biology.

Plant phenotyping relevance

イネ科花序の形態を画像から定量化するPythonベースの表現型解析パイプラインを開発・検証しており、植物表現型の取得・抽出が研究の中心である。

Code and data availability

The paper's SpykProps Python pipeline is openly available on GitHub, and the original/processed spike images, data files, and analysis code are deposited in the University of Minnesota DRUM repository. Both are paper-specific, public, and actionable.