Hello,

We have some updates that you may find helpful in generating predictions for the T3 Prediction Challenge.  We have recently added a large set of phenotyping trials to T3/Wheat and have started developing an R package with some helper functions that you may find useful for programatically fetching data from T3 in R.

WAVE Dataset Added

We have recently added about 2800 trials from the Western Agricultural Variety Explorer (WAVE) Database to T3/Wheat.  These trials are located across 75 locations, primarily in WA, OR, and ID and could be useful as a training set for the Washington state trial in the prediction challenge.  All of the trials can be found under the WAVE breeding program, and are further broken down by region into folders / experiments.  All of the phenotype data for these trials is now available on T3/Wheat.

R Helper Functions

We have developed two straightforward functions that may be helpful for the 2026 T3 Predictathon.  These functions are available as part of the T3 BrAPI Helpers R package, available on GitHub.

First, given a trial that you want to predict, a natural place to look for training data is in other trials where some of the same accessions were evaluated.  Those other trials are likely to have evaluated related germplasm that has similar adaptation.  The T3BrapiHelpers function find_other_studies_evaluating_same_germplasm executes the search for such other trials simply: give the function the studyDbId of a focal trial, a BrAPI Connection from the BrAPI package, and a minimum number of accessions that must be shared, and it will return the set of trials with at least that number in common with the focal trial.

Second, entries to the Predictathon require a specific file structure (described in the rules).  This structure will make it easy for the T3 team to calculate the correlations between observed phenotypes and predictions that will determine the winners of the competition.  To simplify creating the file structure, the T3BrapiHelpers function make_predictathon_file_structure takes a prediction function that you provide and calls that function for each trial of the Predictathon and slots the results into the right structure with the right trial names.

There is of course no need to use either of these functions.  Nevertheless, we hope they will be helpful.

- The Triticeae Toolbox