VenusX: Unlocking Fine-Grained Functional Understanding of Proteins
Yang Tan, Wenrui Gou, Bozitao Zhong, Liang Hong, Huiqun Yu, Bingxin Zhou
TL;DR
VenusX addresses the need for fine-grained protein function understanding by introducing the first large-scale benchmark that spans residue-, fragment-, and domain-level annotations across three task types: residue-level binary classification, fragment-level multi-class classification, and pairwise similarity scoring. It aggregates over 878k samples from InterPro, BioLiP, and SAbDab into 56 datasets, with mix-family and cross-family splits at 50/70/90% sequence identity to probe in-distribution and out-of-distribution generalization. A broad spectrum of baselines, including protein language models, sequence–structure hybrids, structure-based networks, and alignment-based methods, reveal that strong protein-level performance does not guarantee fine-grained functional understanding, underscoring a need for models that capture localized signals and robust generalization. VenusX, with publicly available data and code, provides a biologically meaningful platform to advance function-driven protein representation learning and has potential impact on enzyme design, drug discovery, and structural proteomics, while acknowledging limitations related to scope and evaluation metrics. The benchmark thus sets a foundation for more interpretable, fine-grained modeling of protein function and encourages future research to balance accuracy with localization and biological plausibility.
Abstract
Deep learning models have driven significant progress in predicting protein function and interactions at the protein level. While these advancements have been invaluable for many biological applications such as enzyme engineering and function annotation, a more detailed perspective is essential for understanding protein functional mechanisms and evaluating the biological knowledge captured by models. To address this demand, we introduce VenusX, the first large-scale benchmark for fine-grained functional annotation and function-based protein pairing at the residue, fragment, and domain levels. VenusX comprises three major task categories across six types of annotations, including residue-level binary classification, fragment-level multi-class classification, and pairwise functional similarity scoring for identifying critical active sites, binding sites, conserved sites, motifs, domains, and epitopes. The benchmark features over 878,000 samples curated from major open-source databases such as InterPro, BioLiP, and SAbDab. By providing mixed-family and cross-family splits at three sequence identity thresholds, our benchmark enables a comprehensive assessment of model performance on both in-distribution and out-of-distribution scenarios. For baseline evaluation, we assess a diverse set of popular and open-source models, including pre-trained protein language models, sequence-structure hybrids, structure-based methods, and alignment-based techniques. Their performance is reported across all benchmark datasets and evaluation settings using multiple metrics, offering a thorough comparison and a strong foundation for future research. Code and data are publicly available at https://github.com/ai4protein/VenusX.