PINPOINT: Protease INhibitor PredictiOn at the plant-pathogen INTerface using protein language models and structural modeling

Abstract

Cysteine and serine proteases act as an immune hub in the plant apoplast to provide robust extracellular immunity during microbial colonisation. Microbial pathogens counteract these immune proteases by inhibiting their activity using small secreted proteins (SSPs). Traditionally, SSPs with protease-inhibitory activity are predicted using sequence-dependent database searches. However, in recent years, fungal SSPs have been shown to exhibit protease-inhibitory functions despite lacking the inhibitor domain that is annotated through sequence similarity searches. Hence, a large number of these novel SSPs with putative protease inhibitor functions are missed during detection and filtered out during sequence similarity searches. This necessitates the development of newer approaches to predict SSPs lacking an annotated inhibitor domain. Machine learning approaches, such as protein language models, have emerged as powerful tools for predicting protein functions. To date, no machine learning models have been developed to predict the protease-inhibitory activities of SSPs lacking an annotated inhibitor domain. Here, we introduce a protease inhibitor prediction pipeline, PINPOINT (Protease INhibitor PredictiOn at plant-pathogen INTerface). The PINPOINT pipeline combines fine-tuned protein language model classifiers, a structure-aware autoencoder, and effector prediction into a multi-level framework for identifying SSPs with predicted protease inhibitor functions. PINPOINT predicts protease inhibitors using SSPs sequences and monomeric structures with pre-computed structures obtained from the AlphaFold Protein Structure Database or predicted using the ESMFold public API. We successfully validated the PINPOINT platform using SSPs from the plant fungal pathogen Macrophomina phaseolina. Notably, the PINPOINT platform robustly predicted several of these SSPs as protease inhibitors including Sequence-unrelated but structurally similar (SUSS) effectors. We further validated the inhibitory potential of these predicted M. phaseolina SSPs using AlphaFold Multimer (AFM) screening against candidate apoplastic soybean cysteine and serine proteases. Additionally, this platform can be used as a pre-filtering step in AFM screening approaches to reduce the number of candidates for discovering novel SSPs with protease inhibitor function for cross-kingdom plant-microbe interaction studies. The PINPOINT platform will accelerate the prediction of novel SSPs including SUSS effectors with protease inhibitor functions in proteomes of any organisms. We made the PINPOINT pipeline accessible to the research community as a web-based notebook environment for interactive computing in Google Colab, available at https://github.com/iitj-mpg-lab/PINPOINT