Protein Structure Quiz: From Sequence to 3D Shape

The Protein Data Bank is the single global archive for experimentally determined macromolecular structures and contains over 200,000 structures. X-ray crystallography has historically contributed the largest fraction of entries and provides atomic-resolution electron density maps. Cryo-electron microscopy has grown rapidly in contribution, particularly for large complexes. NMR spectroscopy provides solution-state structures, especially valuable for flexible or membrane-associated proteins.

Comparative homology modeling exploits the principle that proteins with similar sequences adopt similar three-dimensional folds. A template structure from the Protein Data Bank is identified through sequence similarity searching, the target sequence is aligned to the template, and a model is built by copying the backbone coordinates of conserved regions and modeling structurally variable loops. Reliability increases substantially above 30 percent sequence identity and is considered high-quality above 50 percent identity.

AlphaFold2 represents a transformative advance in protein structure prediction. Its Evoformer neural network architecture processes multiple sequence alignment data to extract co-evolutionary information reflecting residue-residue contacts, combined with structural representations refined through attention mechanisms. AlphaFold2 achieves median backbone accuracy comparable to experimental structures for well-represented protein families, producing structures with predicted local distance difference test scores that correlate strongly with actual accuracy, enabling confident structural modeling even for proteins without close homologs.

The Ramachandran plot displays the phi and psi backbone dihedral angles for each non-glycine, non-proline residue in a protein structure. Residues fall into favored regions corresponding to alpha helix, beta strand, and other allowed conformations based on steric energy calculations. Residues falling in disallowed regions may indicate errors in structure determination or modeling. The percentage of residues in favored regions is a standard metric for assessing protein model quality.

Fold recognition or threading addresses the challenge of modeling proteins with no detectable sequence similarity to any known structure, the so-called twilight zone below approximately 20 percent identity. Threading algorithms evaluate how well the target sequence fits the backbone geometry and contact patterns of each known fold template using scoring functions that incorporate sequence-structure compatibility, predicted secondary structure matching, and statistical potentials. It extends structural prediction capability beyond the range accessible to sequence-based homology modeling.

The B-factor in a crystal structure encodes information about how precisely each atom's position is defined by the electron density data. Atoms with high B-factors contribute broad, diffuse electron density, reflecting either genuine thermal motion or multiple alternate conformations in the crystal lattice. High B-factors are commonly observed in surface loops, flexible termini, and active site residues. Bioinformaticians use B-factor profiles to identify flexible or dynamic regions relevant to protein function and drug binding.

Molecular docking computationally samples the conformational and positional space of a ligand within a protein binding site, scoring each pose using energy functions that estimate binding affinity based on electrostatic complementarity, van der Waals contacts, and desolvation effects. Docking enables rapid virtual screening of thousands to millions of compounds against a target structure, prioritizing candidates for experimental testing and reducing the cost and time of early-stage drug discovery.

DALI compares the internal distance geometry of protein structures rather than their sequences. By computing pairwise distances between alpha carbon atoms and comparing these distance matrices, it identifies proteins with similar three-dimensional arrangements of secondary structure elements even when sequence identity is negligible. This approach has revealed numerous examples of structural homologs with shared evolutionary origins and similar active site architectures that are completely invisible to sequence-based methods.

Structural model quality must be rigorously evaluated before using it for drug design, as errors in the binding site region would lead to unreliable docking results. Tools such as PROCHECK assess backbone geometry, Ramachandran plot statistics, and bond length deviations. MolProbity evaluates all-atom clashscores and rotamer quality. The binding pocket region should be specifically examined for alignment quality and template structural coverage, as poorly modeled loops or misaligned residues directly compromise docking reliability.

Cryo-electron microscopy flash-freezes protein samples in vitreous ice, preserving them in a near-native hydrated state. Transmission electron microscopes collect thousands to millions of two-dimensional particle projection images that are computationally combined into a three-dimensional reconstruction. The resolution revolution in cryo-EM since 2013, driven by direct electron detectors and improved software, has enabled routine atomic-resolution structures of large complexes, membrane proteins, ribosomes, viruses, and flexible multi-domain assemblies intractable to crystallization.

When two residues covary across a deep multiple sequence alignment, mutations at one position are compensated by correlated changes at the other to maintain protein stability or function. This pattern implies that the two residues are in direct physical contact or functional communication in the folded structure. Extracting thousands of such covariation signals from large alignments generates a rich set of distance constraints. AlphaFold2 exploits these constraints through its Evoformer attention mechanism to predict accurate inter-residue distance distributions and build high-quality structural models.

Computational model validation requires multiple complementary approaches. Mutagenesis data tests whether residues predicted to be functionally important match experimental findings. Stereochemical quality tools identify backbone and side-chain geometry violations suggesting modeling errors. Experimental methods including limited proteolysis and hydrogen-deuterium exchange provide structural topology and flexibility information that can corroborate or challenge model predictions. Database deposition is a scientific communication step that follows validation, not a prerequisite for it.

Root mean square deviation quantifies structural similarity by calculating the average distance between equivalent atoms, typically alpha carbons, in two protein structures after optimal superimposition using algorithms such as the Kabsch algorithm. Low RMSD values indicate high structural similarity or small conformational changes, while high values reflect significant structural divergence. RMSD is used to compare homologous protein structures, track conformational changes in molecular dynamics simulations, and assess model quality by comparing predictions to experimental structures.

Structural superimposition of distantly related enzymes frequently reveals conserved arrangements of catalytic residues even when overall sequence identity falls below detectable levels. This conservation reflects intense purifying selection on active site geometry because the precise spatial arrangement of catalytic groups is essential for efficient catalysis. Comparing active site architectures across structural homologs reveals mechanistic relationships, informs understanding of enzyme evolution, and provides a rational basis for engineering novel catalytic activities in biotechnology applications.

Virtual screening hit rates of 1 to 10 percent are typical in structure-based drug discovery because docking scoring functions are simplified approximations of binding free energy. They often neglect protein flexibility, accurate solvation, entropic penalties, and quantum mechanical effects. False positives arise when compounds score well geometrically but fail to form stable complexes in solution due to factors the scoring function cannot capture. Despite this limitation, virtual screening efficiently enriches compound collections relative to random screening and remains a valuable early-stage prioritization tool.