Proteome measurements from genetic diversity panels offer the opportunity to robustly classify and predict traits in complex populations. At the same time, the analysis of the proteomes arising from these populations present specific challenges. One challenge lies in establishing a pan-proteome database and/or search strategy such that sequence variation can be robustly identified across genetic diversity. Another challenge lies in establishing a unified set of protein coordinates with which to map all samples’ measurements such that quantitative data are directly comparable between samples. While inroads have been made that seek to meet these challenges, a complete solution has not yet become mainstream. In this regard, approaches to this problem that uncouple the requirement for peptide sequence assignment / protein attribution from classification and prediction offer an alternative solution.
Wheat varieties from the OzWheat genetic diversity panel were grown in glasshouses under simulated long and short daylight conditions using a randomised planting scheme. Samples were taken at the two-leaf stage while sibling plants were grown to full term to capture trait information, such as grain yield, flowering time and spike length. Proteomes were measured for a subset of the cultivars using SWATH acquisition and peak area information extracted using PeakView software. The resulting data were used to classify samples by day length and predict flowering time using a selection of machine learning methods. Recent work has focussed on the conversion of raw SWATH data to images with the goal of applying image classification methods, such as convolution neural networks, for determining day length and predicting yield traits without prior knowledge of peptide/protein coordinates. This image classification work is ongoing at the time of abstract submission with progress to be presented at the conference.