P-174 - Hunting Avrs with AlphaFold-Multimer

Approximately 20% of global wheat production is lost to pests and pathogens every year. Resistance genes are one of the most effective and sustainable approaches for mitigating crop disease. However, in wheat, only a small fraction of HR-inducing effector proteins corresponding to cloned resistance genes have been identified. The aim of this project is to design an ML-assisted screening workflow to identify Avr candidates for a given resistance gene from a list of thousands of putative pathogenic effectors. To develop this workflow, we used as a positive control the stem rust resistance gene Sr62TK along with its known AvrSr62 effector, including its four HR-inducing family members. We used an in-house AlphaFold-Multimer pipeline to generate 7,772 multimeric models of Sr62TK in different stoichiometries against 1,943 putative effectors from the stem rust fungus Puccinia graminis f. sp. tritici. Filtering of these effectors based on structural features and the quality of the binding interfaces reduced this list to 145 proteins, and structural clustering of the remaining models grouped them into 25 distinct clusters. The four AvrSr62 proteins were placed in the same cluster, which ranked number 8 based on cluster size and ipTM score. Overall, our workflow was able to produce a shortlist of ~25 proteins containing the true AvrSr62, from a starting list of 1,943 putative effectors. This approach promises to greatly accelerate the identification of Avr proteins and their study.