PIONEER (Protein-protein InteractiON IntErfacE pRediction) is a deep learning-based ensemble learning pipeline for accurate partner-specific protein interface prediction. Compared to other published state-of-the-art methods, including our previously developed method - ECLAIR, PIONEER has shown remarkable advantages in both quality and coverage. PIONEER is an ensemble model of two advanced deep learning architectures: graph convolutional neural networks with auto-regressive moving average filters and bidirectional recurrent neural networks with gated recurrent units. These two models are suited to proteins with and without structure information, respectively, and their integration provides the highest performance for protein interface prediction. For each unresolved protein interaction, PIONEER leverages the maximum structure and sequence information (solvent-accessible surface area, molecular docking results, biophysical features, evolutionary conservation, co-evolutionary features, RaptorX-predicted structure features, and pair-potential features). To make the most accurate interface predictions, PIONEER utilizes all possible structure information (PDB, Modbase models, and high-quality AlphaFold2-predicted structures) in addition to the comprehensive residue features. PIONEER has already predicted interfaces for 256,946 protein interactions without structure information in humans and seven other commonly studied organisms (A. thaliana, C. elegans, D. melanogaster, E. coli, M. musculus, S. cerevisiae, and S. pombe). The prediction results are readily available under the Download tab. Users can also submit any protein interactions for interface prediction. Moreover, this website helps users explore protein structures, protein interactions and their associated diseases.


Version 1.1 January 2024


The PIONEER Interactive Browser is provided by the Yu lab at Cornell University. Please contact us at for any issues.