{
  "id": "protein-secondary-structure",
  "title": "Protein Secondary Structure",
  "lang": "en",
  "source_lang": "en",
  "summary": "Protein secondary structure comprises recurring local backbone conformations, including helices, sheets, and turns, that help organize protein architecture and function.",
  "category": "science",
  "subcategory": "biology",
  "type": "concept",
  "aliases": [
    "Secondary structure of proteins",
    "Polypeptide secondary structure"
  ],
  "generated_at": "2026-11-04",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/protein-secondary-structure/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/protein-secondary-structure/",
      "json": "https://aiwiki.page/zh/protein-secondary-structure.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Protein secondary structure is the recurring spatial organization of the backbone of a protein, most prominently into α-helices and β-sheets. These arrangements arise from the geometry of the polypeptide chain and from interactions among its chemical groups, particularly backbone hydrogen bonds. Secondary structure describes an intermediate level of organization between the amino-acid sequence and the complete three-dimensional fold, although its formation is closely coupled to the packing of the whole molecule.\n\nThe concept is central to structural biology because a limited repertoire of backbone conformations supports an enormous diversity of protein shapes and functions. Helices can span membranes, sheets can form stable molecular frameworks, and turns can redirect a chain or position functional groups. Secondary structure is investigated through atomic structures, spectroscopy, physical models, and computational prediction. Its boundaries are not always unique: different methods emphasize hydrogen-bond patterns, backbone angles, or experimental signals, and some protein regions occupy changing ensembles rather than a single conformation."
    },
    {
      "heading": "Backbone geometry and the meaning of secondary structure",
      "text": "A polypeptide consists of amino-acid residues connected through peptide bonds. Its repeating backbone contains an amide nitrogen, an α-carbon, and a carbonyl carbon. Electron delocalization gives the peptide carbon–nitrogen bond substantial partial double-bond character, so the atoms of each peptide group are approximately coplanar. Most peptide bonds adopt the trans configuration, with the adjacent α-carbons on opposite sides of the peptide bond. Cis peptide bonds are uncommon but occur more frequently before proline than before other residues. This restricted geometry sharply reduces the conformations available to a chain.\n\nMost remaining backbone flexibility is described by two torsion angles per residue: \\(\\phi\\), around the nitrogen–α-carbon bond, and \\(\\psi\\), around the α-carbon–carbonyl-carbon bond. The peptide-bond torsion angle is designated \\(\\omega\\). Not every combination of \\(\\phi\\) and \\(\\psi\\) is physically accessible, because some would bring nonbonded atoms too close together. A Ramachandran plot displays these angular combinations and identifies regions compatible with common conformations. Right-handed α-helices typically occupy a region near \\(\\phi=-60^\\circ\\), \\(\\psi=-45^\\circ\\), whereas extended β-strand conformations occupy a broad region with negative \\(\\phi\\) and positive \\(\\psi\\). These are distributions rather than exact universal coordinates.\n\nBackbone hydrogen bonds usually involve a carbonyl oxygen as acceptor and an amide N–H group as donor. Repetition of suitable donor–acceptor arrangements produces characteristic helical or sheet patterns. Proline is distinctive because its side chain closes onto the backbone nitrogen: an internal proline residue has no backbone N–H donor, and its ring restricts \\(\\phi\\). Glycine, whose side chain is only hydrogen, has unusually broad conformational freedom. These chemical differences help explain why particular residues favor or disrupt particular structural environments.\n\nSecondary structure is often called “local structure,” but this expression requires qualification. An α-helix is stabilized largely by interactions among residues close together in sequence. In a β-sheet, neighboring strands can come from widely separated parts of a chain or even from different chains. What makes the sheet secondary structure is its recurring backbone geometry and hydrogen-bond organization, not strictly short sequence separation. By contrast, tertiary structure describes the overall arrangement of a single chain, while quaternary structure describes the organization of multiple protein subunits. Side-chain packing, strand connectivity, and the relative positions of helices usually belong to these higher levels."
    },
    {
      "heading": "Helices and their structural variants",
      "text": "The α-helix is the most common regular helical conformation in proteins. In an ideal α-helix, the carbonyl group of residue \\(i\\) hydrogen-bonds to the amide group of residue \\(i+4\\). Approximately 3.6 residues complete one turn, each residue advances about \\(1.5\\) Å along the axis, and the pitch is about \\(5.4\\) Å per turn. These ideal values vary slightly in actual structures. Because proteins are built predominantly from L-amino acids, their α-helices are overwhelmingly right-handed. Side chains project outward from the helical backbone, leaving the central region densely packed rather than hollow.\n\nThe helical hydrogen-bond pattern cannot be completed at either end. Several backbone donors or acceptors near the termini therefore interact with water, side chains, or other parts of the protein. Specific capping arrangements help satisfy these groups and favor a particular helix boundary. The approximately aligned peptide-group dipoles also produce a helix macrodipole, with a partially positive amino-terminal end and a partially negative carboxyl-terminal end. Nearby charged groups can interact with this electrostatic pattern, although their effects depend strongly on solvent exposure and the surrounding structure.\n\nDifferent helical geometries have different hydrogen-bond registers. A \\(3_{10}\\) helix has \\(i\\)-to-\\(i+3\\) hydrogen bonds and about three residues per turn; its axial rise is approximately \\(2.0\\) Å per residue. Such helices are usually short and often occur at α-helix termini. A π-helix has \\(i\\)-to-\\(i+5\\) hydrogen bonds and roughly 4.4 residues per turn. π-helical segments are less common and frequently appear as short expansions within otherwise α-helical regions. The traditional notation \\(3_{10}\\) refers to three residues per turn and a ten-atom hydrogen-bonded ring, not to a decimal number.\n\nHelices need not be straight or uninterrupted. Proline can introduce a kink because of its restricted backbone geometry and missing amide hydrogen, but its presence does not invariably terminate a helix. Glycine can permit bending or close packing. These distortions are especially important in membrane proteins, where small changes in helical orientation can alter an internal cavity or open an ion-conducting pathway. An amphipathic helix places hydrophobic and hydrophilic residues on different faces; its properties depend on the periodic placement of residues around the helical circumference.\n\nSeveral α-helices can wind around one another to form a coiled coil. A common sequence pattern is a seven-residue repeat, conventionally labeled \\(a\\) through \\(g\\), with hydrophobic residues frequently occupying positions \\(a\\) and \\(d\\). Their packing helps stabilize the superhelical assembly. Coiled coils occur in structural and regulatory proteins, but the coiled-coil arrangement itself is a higher-order association of helices rather than a separate backbone conformation."
    },
    {
      "heading": "β-strands, β-sheets, and extended assemblies",
      "text": "A β-sheet is formed when extended segments of polypeptide, called β-strands, associate through backbone hydrogen bonds. The backbone of an individual strand follows a pleated path rather than a straight line, and successive side chains generally project to alternating sides. The advance along an extended strand is approximately \\(3.3\\)–\\(3.5\\) Å per residue, considerably greater than the axial advance per residue in an α-helix. A β-strand is therefore a local backbone conformation, whereas a β-sheet also specifies an association between strands.\n\nSheets are classified according to strand direction. In an antiparallel sheet, adjacent strands run in opposite amino-to-carboxyl directions; in a parallel sheet, they run in the same direction. Mixed sheets contain both relationships. Antiparallel sheets generally permit relatively direct hydrogen bonds between paired residues, whereas parallel sheets have an offset bonding pattern in which a residue interacts with different positions on the neighboring strand. Both arrangements are widespread, and neither is universally more stable: stability also depends on sequence, sheet size, solvent exposure, and packing against other structural elements.\n\nReal β-sheets are usually twisted rather than flat. The predominant twist reflects the stereochemistry of L-amino-acid backbones and the energetics of their packing. Sheets can curve into barrels or pack face-to-face as β-sandwiches. An antiparallel β-barrel can close through hydrogen bonding between its first and last strands, while larger architectures can incorporate several sheets and connecting loops. Local irregularities include β-bulges, in which unequal numbers of residues on neighboring strands disturb the usual register. Such features can change sheet curvature, accommodate an insertion, or redirect a chain.\n\nThe alternating orientation of side chains gives β-strands a useful sequence periodicity. Alternating polar and nonpolar residues can produce one face suited to water exposure and another suited to a hydrophobic core. However, this is a statistical tendency, not a definition: many strands lack simple alternating patterns. Edge strands present a particular structural problem because they expose potential backbone hydrogen-bonding groups. Proteins can limit unwanted sheet extension through charged residues, prolines, irregular geometry, protective loops, or packing against other structural elements.\n\nβ-sheet architecture also extends beyond individual folded proteins. In amyloid fibrils, many polypeptide molecules assemble into cross-β structures, with β-strands running approximately perpendicular to the fibril axis and sheets extending along it. A characteristic spacing of about \\(4.7\\) Å corresponds to neighboring strands. Amyloid demonstrates that the same backbone chemistry can support both a finite native fold and an extended intermolecular assembly. The resulting fibril structures are not all identical: their folds, side-chain contacts, and numbers of protofilaments differ among proteins and can differ among assemblies of the same sequence."
    },
    {
      "heading": "Turns, loops, and noncanonical conformations",
      "text": "Helices and sheets do not exhaust the structural possibilities of a polypeptide. A chain must frequently change direction, connect distant structural elements, or position a surface segment without adopting a long repetitive pattern. Turns and loops perform these roles. A turn is a relatively compact directional change with a recognizable local geometry; a loop is a broader descriptive category for a connecting segment. The terms overlap, and a loop may contain one or more turns, short helices, or extended residues. Neither term necessarily implies mobility or structural disorder.\n\nA β-turn usually involves four consecutive residues, conventionally numbered \\(i\\) through \\(i+3\\), and often includes a hydrogen bond from the carbonyl of \\(i\\) to the amide group of \\(i+3\\). Geometric definitions can identify β-turns even when this hydrogen bond is absent. Different turn types are distinguished chiefly by the \\(\\phi\\) and \\(\\psi\\) angles of the two central residues. Types I and II are common examples, with corresponding “prime” types occupying approximately mirror-related angular regions. Glycine is favored at positions requiring positive \\(\\phi\\), particularly the third residue of a conventional type II turn, while proline is frequent at positions compatible with its restricted backbone angle.\n\nShorter γ-turns involve three residues and can contain an \\(i\\)-to-\\(i+2\\) hydrogen bond. Longer chain reversals include several geometrically distinct families. A β-hairpin consists of two antiparallel β-strands connected by a short reversal, often but not always a conventional β-turn. It combines a local chain connection with strand–strand hydrogen bonding. Hairpin geometry influences how a sheet grows and how residues become paired across it, making the turn region important to both folding and final architecture.\n\nThe polyproline II helix is another important backbone conformation. It is an extended, usually left-handed helix with approximately three residues per turn and predominantly trans peptide bonds. It does not require a repeating network of intrachain backbone hydrogen bonds and is not restricted to proline-only sequences. Polyproline II-like conformations occur in exposed segments and in peptides recognized by protein-binding domains. In collagen, three chains with related left-handed conformations associate into a right-handed triple helix. Its repeating glycine-rich sequence permits exceptionally close packing near the triple-helix axis.\n\nIrregular structure must also be distinguished from intrinsic disorder. A well-resolved active-site loop can have a reproducible, highly specific shape without resembling any canonical helix or strand. Conversely, an intrinsically disordered protein or region samples an ensemble of conformations rather than maintaining a single stable fold under the conditions being considered. Such ensembles may contain transient helices, turns, or extended conformations. Describing all nonhelical, nonsheet residues as “random coil” therefore collapses several physically different situations into one convenient but potentially misleading category."
    },
    {
      "heading": "Energetics, sequence preferences, and folding",
      "text": "The stability of secondary structure is a balance among several energetic and entropic contributions. Forming a helix or sheet restricts backbone conformational freedom, which generally carries an entropy cost. Favorable hydrogen bonds, electrostatic interactions, side-chain contacts, and interactions with the solvent can compensate for that cost. A backbone hydrogen bond is not an isolated gain added to an otherwise unchanged system: before folding, its donor and acceptor may already hydrogen-bond to water. The relevant quantity is the free-energy difference between alternative molecular and solvent arrangements.\n\nAt constant temperature and pressure, this balance is commonly expressed through Gibbs free energy:\n\n\\[\\Delta G=\\Delta H-T\\Delta S.\\]\n\nHere, \\(\\Delta H\\) includes changes in interactions within the chain and with its environment, while \\(\\Delta S\\) includes changes in chain and solvent freedom. The hydrophobic effect favors burial of many nonpolar groups in aqueous proteins and thereby helps organize the packing that stabilizes helices and sheets. Once a backbone segment is buried, leaving its polar groups without suitable interaction partners can be costly. Regular secondary structures provide efficient ways to satisfy many of those groups.\n\nAmino acids have measurable conformational preferences, but these depend on context. Alanine often favors α-helical conformations in model peptides. β-branched residues such as valine and isoleucine are frequently compatible with β-strands. Glycine can destabilize a helix because its unfolded state has especially high conformational freedom, while proline can interrupt a regular helical hydrogen-bond network. These tendencies are not deterministic rules. Neighboring residues, terminal capping, burial, interactions with other parts of the protein, and chemical conditions can override a preference measured in an isolated peptide.\n\nHelix formation illustrates cooperativity. Creating an initial ordered segment incurs a substantial conformational cost, but extending an existing helix can be more favorable because additional residues join an established hydrogen-bonding arrangement. Statistical models of the helix–coil transition, including the Zimm–Bragg model introduced in 1959, distinguish nucleation from propagation. Sheet formation has additional complications because strands must find the correct partners, orientation, and residue register. A sequence may support several plausible strand pairings, only one of which belongs to its final native fold.\n\nDuring protein folding, secondary structure may form early, develop together with long-range packing, or reorganize as folding proceeds. There is no universal sequence in which all proteins first complete helices and sheets and only afterward assemble tertiary structure. Likewise, loss of activity during unfolding need not coincide exactly with loss of secondary structure. Partly folded intermediates can retain substantial helical or sheet content while lacking tightly organized side-chain packing. Temperature, pH, solvent composition, and binding partners can shift these equilibria, so structural descriptions refer implicitly to particular experimental or cellular conditions."
    },
    {
      "heading": "Structural roles in protein function",
      "text": "Secondary structures provide mechanically and chemically distinct building elements. α-Helices form compact rods whose side chains can create polar, nonpolar, or charged surfaces. β-Sheets provide extended surfaces and frameworks that can bend, twist, or close into barrels. Loops place functional groups in locations not easily reached by repetitive backbones. A protein domain often combines these elements into a compact unit capable of folding or functioning relatively independently. Domain function nevertheless depends on the detailed arrangement and chemistry of its residues, not simply on its percentage of helix or sheet.\n\nMembrane proteins show especially clear relationships between secondary structure and environment. In an α-helical transmembrane segment, most backbone donors and acceptors participate in internal hydrogen bonds, while outward-facing side chains contact the membrane interior. Many transmembrane helices contain roughly 20–25 residues, although the required length varies with helix tilt, membrane thickness, and local deformation. G-protein-coupled receptors characteristically contain seven transmembrane helices. In contrast, many outer-membrane proteins of bacteria use β-barrels, with one set of strand side chains facing the lipid environment and another facing the barrel interior.\n\nIn soluble proteins, secondary structures organize catalytic and binding sites. The α/β barrel found in many enzymes consists of eight parallel β-strands surrounded by α-helices; catalytic residues are often positioned in loops near one end of the barrel. The immunoglobulin fold, used by antibodies and many other proteins, is based on a β-sandwich. Antibody recognition depends strongly on variable loops supported by this framework. In hemoglobin, predominantly α-helical globin subunits create pockets that hold heme and support reversible oxygen binding. These examples demonstrate that a structural framework supports function without uniquely determining it.\n\nSecondary structure also affects mechanical behavior. Pulling a β-sheet in a direction that loads several hydrogen bonds together can require a different force from peeling strands apart sequentially. Helices can extend through progressive disruption of their hydrogen-bond network, while coiled coils can accommodate deformation through changes in packing and superhelical geometry. Such behavior depends on loading direction, sequence, solvent conditions, and pulling rate. The distinction is important in structural proteins and in proteins exposed to mechanical stress, where molecular architecture governs how force is transmitted through the chain.\n\nChanges in secondary-structure geometry can contribute to molecular switching without requiring wholesale unfolding. A helix may tilt, bend, rotate, or change its interactions with neighboring helices; a loop may become ordered after ligand binding; a strand may exchange partners during assembly. Some proteins undergo larger fold changes in which the same sequence supports substantially different secondary-structure arrangements. Conversely, closely related proteins can retain similar helices and sheets despite considerable sequence divergence. Secondary structure is therefore a useful level for comparing proteins, but functional interpretation also requires connectivity, dynamics, and the molecular environment."
    },
    {
      "heading": "Experimental determination and spectroscopic signatures",
      "text": "Atomic-resolution methods reveal secondary structure through the positions and organization of backbone atoms. In X-ray crystallography, a structural model is built and refined against diffraction data, allowing backbone angles and likely hydrogen-bond patterns to be evaluated. Hydrogen atoms are usually not directly resolved in routine protein structures, so hydrogen bonds are commonly inferred from donor–acceptor geometry. Poorly ordered regions may lack interpretable density, and an absent segment should not automatically be assigned a particular conformation. Resolution, model quality, crystal contacts, and alternative conformations all affect interpretation.\n\nNuclear magnetic resonance provides complementary information in solution and, with specialized methods, in solids. Nuclear Overhauser effects report spatial proximity between nuclei; characteristic short- and medium-range patterns help identify helices, while interstrand contacts support sheet arrangements. Backbone chemical shifts are sensitive to local conformation. Differences between observed shifts and suitable reference values, often called secondary chemical shifts, provide information about helical or extended tendencies. Scalar couplings constrain torsion angles, and exchange behavior can reveal protection of backbone amide hydrogens, although protection is not determined by secondary structure alone.\n\nCryo-electron microscopy can identify helices and sheets within three-dimensional reconstructions of large molecular assemblies. At sufficiently high resolution, individual strands and backbone features become distinguishable; at lower resolution, helices may appear as rods while sheets are less precisely resolved. Interpretation combines experimental density with chemically plausible molecular models. None of these methods eliminates structural uncertainty: local resolution and conformational variability can differ markedly across the same molecule, so an apparently precise atomic model must still be evaluated against its supporting data.\n\nCircular dichroism measures the differential absorption of left- and right-circularly polarized light. In the far-ultraviolet region, peptide electronic transitions produce signals sensitive to backbone conformation. α-Helical proteins commonly show negative bands near 208 and 222 nm and a positive band near 190 nm. β-rich proteins often show a negative band around 215–218 nm and a positive band near 195 nm, but their spectra are especially variable. Reference-based analysis can estimate structural fractions, yet these estimates depend on spectral range, sample quality, concentration, and the suitability of the reference set. A spectrum does not directly identify which individual residues form a helix.\n\nInfrared spectroscopy and Raman spectroscopy probe molecular vibrations. The amide I region, approximately \\(1600\\)–\\(1700\\ \\mathrm{cm}^{-1}\\), is dominated by peptide carbonyl stretching and is particularly sensitive to hydrogen bonding and coupling among peptide groups. Overlapping bands can be analyzed to estimate structural components, including changes during aggregation or unfolding. Solvent absorption, side-chain contributions, and fitting assumptions complicate interpretation. Spectroscopic estimates are therefore most informative when compared with appropriate controls and, where possible, residue-resolved structural measurements rather than treated as uniquely determined structural maps."
    },
    {
      "heading": "Structure assignment, prediction, and computational analysis",
      "text": "Even with atomic coordinates available, assigning a secondary-structure label requires a definition. The widely used DSSP method, introduced by Wolfgang Kabsch and Christian Sander in 1983, recognizes patterns of backbone hydrogen bonding using an electrostatic model. Its conventional detailed states distinguish α-helices, \\(3_{10}\\) helices, π-helices, extended β-strands, isolated β-bridges, turns, bends, and residues not assigned to these classes. A bend or an unassigned position is not necessarily disordered. DSSP labels summarize a geometric model; they are not separate observations made directly by an instrument.\n\nOther algorithms combine backbone torsion angles with hydrogen bonds or use different geometric criteria. Consequently, the first or last residue of a helix may receive different labels in different assignments, as may distorted strands, short helices, and turn regions. Simplified three-state classifications usually group residues into helix, strand, and coil, but the exact mapping of detailed states is not universal. Reported helix or sheet percentages should therefore be interpreted alongside the assignment method, the treatment of missing residues, and whether the denominator includes an entire sequence or only its modeled portion.\n\nPrediction from sequence asks a different question: which conformations are likely before coordinates are known? Early statistical approaches used amino-acid preferences and neighboring sequence patterns. Later methods incorporated evolutionary information from aligned homologous sequences, because substitutions can reveal structural constraints that are not obvious in a single sequence. Machine-learning models then learned relationships between sequence context and assigned structure. Their outputs are often residue-level probabilities rather than unqualified labels, reflecting the fact that some positions have ambiguous boundaries or condition-dependent conformations.\n\nPerformance is commonly measured by three-state per-residue accuracy, often called \\(Q_3\\), but a single number has limitations. Correctly predicting a long helix can compensate numerically for missing a short strand. Segment-based measures assess whether the positions and lengths of structural elements are recovered. Reliable evaluation also requires separation between training and test proteins, especially between closely related sequences. Otherwise, similarity to examples in the training data can make generalization appear stronger than it is. Experimental assignment itself supplies imperfect labels at irregular or poorly resolved positions.\n\nModern three-dimensional prediction systems can supply a complete coordinate model from which secondary structure is subsequently assigned. This does not make local structural analysis redundant: helix and strand annotations remain useful for model comparison, visualization, fold classification, and detecting implausible geometry. Experimental coordinates distributed through the Protein Data Bank support such analyses, while molecular dynamics simulations can track how secondary-structure assignments change through time. In simulations, conclusions depend on the interaction model, sampling, solvent representation, and assignment algorithm. A fluctuating trajectory is often better represented by the fraction of time each residue occupies a state than by one permanent label."
    },
    {
      "heading": "Historical development and changing interpretations",
      "text": "The modern concept developed from the recognition that protein chains possess a regular chemical backbone even when their sequences differ. In the 1930s, William Astbury’s X-ray studies of fibrous proteins distinguished diffraction patterns associated with α- and β-forms, particularly in work on keratin and stretched fibers. These observations established that proteins could adopt reproducible structural organizations and that mechanical treatment could alter them. However, identifying an entire atomic arrangement required more than matching a few characteristic spacings: the geometry had to be compatible with peptide chemistry and stereochemical constraints.\n\nA decisive advance came in 1951, when Linus Pauling, Robert Corey, and Herman Branson described the α-helix using realistic bond geometry, approximately planar peptide groups, and systematic hydrogen bonding. Pauling and Corey also presented β-sheet models that year. Their work showed how a chemically plausible polypeptide backbone could generate repeating structures without assuming an integer number of residues per turn. The α-helix’s approximately 3.6 residues per turn was particularly important because it demonstrated the value of deriving geometry from molecular constraints rather than imposing an overly simple repeat.\n\nSubsequent experiments connected these models to actual proteins. Francis Crick’s analysis of helical diffraction and coiled-coil organization helped interpret the architecture of fibrous proteins. John Kendrew and colleagues reported a low-resolution structure of myoglobin in 1958 and a much more detailed structure in 1960, revealing extensive α-helical organization in a globular protein. Max Perutz and colleagues’ studies of hemoglobin established related structural principles in a multisubunit protein. These results made clear that regular secondary structures could be embedded in irregular, compact three-dimensional folds rather than existing only as long uniform fibers.\n\nIn 1963, G. N. Ramachandran, C. Ramakrishnan, and V. Sasisekharan published a systematic analysis of sterically permitted backbone conformations. The resulting angular framework connected secondary structure to the physical dimensions of atoms and became a foundation for structural validation. As more protein structures became available, researchers documented distributions rather than single ideal geometries, including unusual conformations supported by local interactions. Hydrogen-bond-based assignment methods and public coordinate archives later made comparisons across large collections of proteins practical and reproducible.\n\nThe growing structural record also changed the treatment of regions outside helices and sheets. Early schematic representations often grouped them together as irregular chain. Detailed crystallography, solution measurements, and studies of intrinsically disordered proteins instead revealed a spectrum ranging from precisely organized loops to broad, rapidly exchanging ensembles. Secondary structure consequently remains both a physical concept and an analytical classification. Its most informative use preserves the distinction between idealized recurring patterns, experimentally supported conformations, and the dynamic populations through which a protein moves."
    }
  ],
  "links": [
    "protein",
    "amino-acid",
    "peptide-bond",
    "ramachandran-plot",
    "hydrogen-bond",
    "protein-tertiary-structure",
    "alpha-helix",
    "beta-sheet",
    "amyloid",
    "collagen",
    "entropy",
    "water",
    "gibbs-free-energy",
    "hydrophobic-effect",
    "protein-folding",
    "protein-domain",
    "bacteria",
    "enzyme",
    "antibody",
    "hemoglobin",
    "x-ray-crystallography",
    "nuclear-magnetic-resonance",
    "cryo-electron-microscopy",
    "infrared-spectroscopy",
    "raman-spectroscopy",
    "machine-learning",
    "training-data",
    "protein-data-bank",
    "molecular-dynamics",
    "linus-pauling"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}