Quaternary Structure Prediction

The core computational method for predicting the 3D structure of a complex from the unbound structures of its individual components. Algorithms search the vast conformational space of rigid-body orientations and internal flexibilities.

• Global Docking: Scans the entire surface of both proteins, assuming minimal conformational change.
• Local Docking: Refines an approximate complex structure, often incorporating side-chain and backbone flexibility.
• Scoring Functions: Evaluate the complementarity of interfaces using physics-based energy terms (van der Waals, electrostatics) or statistical potentials derived from known structures.
• CAPRI (Critical Assessment of PRedicted Interactions): The community-wide blind experiment that benchmarks docking algorithms, analogous to CASP for single-chain prediction.

A specialized version of AlphaFold2 developed by DeepMind, trained specifically on protein complexes in the Protein Data Bank. It predicts the structure of multi-chain assemblies from their concatenated sequences.

• Chain Indexing: Distinguishes identical sequences by modifying the residue index, preventing the model from collapsing homomeric chains into a single entity.
• Cross-Chain Attention: The Evoformer and Structure Module process all chains simultaneously, allowing the model to learn inter-chain residue-residue relationships.
• Symmetry Handling: Can accurately predict the structure of symmetrical complexes, such as viral capsids and ion channels, without explicit symmetry constraints.
• Output: Provides a single, unified 3D structure for the entire complex, along with per-residue (pLDDT) and per-residue-pair (PAE) confidence metrics.

Standard confidence metrics must be reinterpreted for quaternary structures. The reliability of a predicted interface is not captured by the average pLDDT of the whole complex.

• Interface pLDDT (ipLDDT): The average pLDDT score calculated exclusively for residues at the binding interface. A high ipLDDT (>80) indicates a confident interface prediction.
• Cross-Chain PAE: The Predicted Aligned Error between residues on different chains. Low cross-chain PAE values (<10 Å) signify high confidence in the relative positioning of subunits.
• DockQ Score: A composite metric that combines interface RMSD, fraction of native contacts, and ligand RMSD to provide a single quality score for a predicted complex, widely used in CAPRI.

Many protein complexes form highly symmetrical oligomers. Exploiting this symmetry is a powerful constraint for prediction and design.

• Cyclic (Cn) Symmetry: Subunits arranged in a ring, common in membrane channels and DNA-binding proteins.
• Dihedral (Dn) Symmetry: Two stacked rings, forming a barrel-like structure.
• Icosahedral Symmetry: The highest-order point group symmetry, used by viral capsids to build large containers from a small number of protein types.
• Symmetry-Constrained Docking: Algorithms like Rosetta SymDock enforce specific symmetry operations during the search, dramatically reducing the computational complexity of assembling large, repetitive structures.

Two fundamental paradigms exist for predicting the structure of complexes where individual subunits are not stably folded in isolation, such as coiled-coils and amyloid fibrils.

• Fold-and-Dock: First predicts the folded monomer structure, then docks the rigid monomers. This fails for obligate complexes where folding and binding are coupled.
• Dock-and-Fold: Simultaneously folds the polypeptide chains and assembles the complex. This is the approach implicitly used by AlphaFold-Multimer.
• Coupled Folding and Binding: A hallmark of many intrinsically disordered proteins (IDPs) that only adopt a stable 3D structure upon binding their partner.

A direct application of quaternary structure prediction for designing vaccines and therapeutic antibodies. It identifies the specific 3D surface patches on an antigen recognized by the immune system.

• Conformational B-Cell Epitopes: Discontinuous patches of amino acids brought together by the protein fold. They cannot be predicted from sequence alone and require a 3D structure.
• Antibody-Antigen Docking: A specialized docking challenge due to the long, flexible CDR loops of antibodies. Tools like Rosetta SnugDock and AlphaFold-Multimer are specifically adapted for this task.
• Epitope Binning: Clustering antibodies into groups that compete for the same binding site, used to rationally design antibody cocktails against rapidly mutating pathogens.