PEACE: Prototype-aware Effector Analysis via Contrastive Embeddings

Abstract

Pathogenic fungi and oomycetes secrete effector proteins that manipulate host defenses and physiology to facilitate infection. However, in a typical secretome, effectors represent only a small fraction of proteins, creating extreme class imbalance and making deep-learning-based effector prediction prone to false positives. Here, we introduce PEACE (Prototype-aware Effector Analysis via Contrastive Embeddings), a lightweight pipeline that integrates ProtTrans (ProtT5) sequence embeddings with prototype-aware contrastive training to enhance effector identification. We benchmark PEACE against EffectorP 3.0 on two datasets with realistic class ratios: a fungal-only dataset and a combined fungi+oomycete dataset. PEACE outperforms EffectorP 3.0 on both datasets, while maintaining highly competitive recall. Post-hoc analysis shows that PEACE forms compact effector clusters against a well-dispersed non-effector background, which improves precision in the high-recall regime. These findings demonstrate that prototype-aware objectives, combined with curated data, can improve effector discovery for high-throughput screening in plant pathology and biotechnology.