Challenges and opportunities in type III secretion system effector prediction

Abstract

Type III secretion system effectors (T3SEs) are small bacterial proteins with big biological roles. They act as central molecular mediators of interactions between Gram-negative bacteria and eukaryotic hosts, spanning pathogenic, symbiotic and environmental contexts. Over the past three decades, T3SE discovery has progressed from genome-independent experimental assays to an expanding landscape of computational prediction methods. Early in silico approaches formalized empirically defined protein N-terminal properties into feature-engineered machine-learning models, followed by deep-learning methods that learn sequence patterns directly from amino acid sequences. More recent pipelines integrate multiple layers of information, including homology, regulatory elements, genomic context, pan-genomic context and protein language model embeddings, primarily to prioritize candidate novel effectors. Despite these advances, several challenges remain. Training data and available databases remain biased towards a limited set of well-known plant and animal pathogens; many tools are no longer maintained, and the extent to which current predictors generalize to non-pathogenic, symbiotic, environmental and host-unknown bacteria remains unclear. Here, we review the conceptual evolution of T3SE prediction, highlight persistent limitations and sources of bias, and outline open questions that must be addressed to enable robust, interpretable and ecologically inclusive prediction of T3SEs, pointing towards the need for centralized, user-friendly platforms that integrate diverse biological signals into transparent, ranked outputs suitable for experimental validation.