Abstract
In the last decade, de novo protein structure prediction accuracy for individual proteins has improved significantly by utilising deep learning (DL) methods for harvesting the co-evolution information from large multiple sequence alignments (MSA). In CASP14, the best groups predicted the structure of most proteins with impressive accuracy. The same approach can, in principle, also be used to extract information about evolutionary-based contacts across protein-protein interfaces. This paper introduces a fold-and-dock method, PconsDock, based on predicted residue-residue distances with trRosetta. PconsDock can simultaneously predict the tertiary and quaternary structure of a protein pair, even when the structures of the monomers are not known. The straightforward application of this method to a standard dataset for protein-protein docking yielded limited success. However, using alternative methods for MSA generating allowed us to dock accurately significantly more proteins. We also introduced a novel scoring function, PconsDock, that accurately separates 98% of correctly and incorrectly folded and docked proteins. The average performance of the method is comparable to the use of traditional, template-based or ab initio shape-complementarity-only docking methods. Moreover, the results of conventional and fold-and-dock approaches are complementary, and thus a combined docking pipeline could increase overall docking success significantly. PconsDock contributed to the best model for one of the CASP14 oligomeric targets, H1065.
Introduction: The Challenge of Protein Docking
In structural biology, "docking" is the process of predicting how protein subunits fit together. Traditionally, this requires solved structures of the individual monomers. However, many proteins lack known structures. While deep learning has revolutionized single-protein folding, protein-protein interactions remain challenging due to the "pairing problem": identifying interacting orthologs across species to extract co-evolutionary signals. PconsDock addresses this by integrating folding and docking into a single pipeline driven by evolutionary data.
Methods: The PconsDock Pipeline
PconsDock utilizes co-evolutionary information to predict inter-chain and intra-chain residue distances simultaneously. The pipeline consists of four main stages:
- Joint MSA Generation: Genomic databases are searched to find likely interacting protein pairs, creating a paired Multiple Sequence Alignment.
- Distance Prediction: The joint MSA is processed by trRosetta to predict distance constraints between all residues.
- Folding and Docking: These constraints guide Rosetta or CNS to physically model the complex from scratch.
- Consensus Scoring: A novel scoring function evaluates multiple models from different alignment strategies to identify the most physically plausible structure.
Key Insights: The Alignment Lottery and MSA Depth
A central finding of this research is the "Alignment Lottery": no single alignment strategy works for all proteins. Some complexes require deep, diverse alignments, while others benefit from smaller, high-precision sets. The depth of the alignment, measured by the number of effective sequences (Meff), strongly correlates with the quality of the predicted structure (TMscore) and the docking accuracy (dockQ).
Furthermore, the authors identified "pseudo-homodimers" as a significant challenge. When interacting proteins are evolutionarily related, the model can mistake internal folding signals for interface contacts, leading to structural artifacts.
Key Results: Benchmarking and Performance
On the Dockground 4.3 benchmark, PconsDock successfully folded and docked approximately 10% of cases from scratch. While this success rate may seem modest, it is comparable to traditional docking methods that start with known monomer structures. Notably, PconsDock often succeeds on targets where shape-based docking fails, suggesting these approaches are complementary.
The consensus scoring method proved exceptionally robust, achieving an AUC > 0.93 in distinguishing correct models from incorrect ones. This high precision allows researchers to trust the system's successful predictions even when the overall success rate is low. The protocol's efficacy was further validated in CASP14, where it contributed to the top-ranked model for target H1065.
Conclusion and Future Directions
PconsDock demonstrates that simultaneous folding and docking using evolutionary information is a viable path forward, especially for complexes with unknown monomer structures. Future work will focus on refining interface geometry, automated selection of optimal alignment strategies, and better handling of paralogs to improve the pairing process.
Summary updated