# Kyte-Doolittle hydropathy scale and how to read a hydropathy plot > The Kyte-Doolittle hydropathy scale gives each amino acid a value from +4.5 (isoleucine) to −4.5 (arginine). Learn all 20 values, how Kyte and Doolittle derived them, how window size shapes a hydropathy plot, where the 19-residue window and 1.6 cutoff come from, how often that rule is right on the human proteome, and how to read peaks, valleys and GRAVY. Paste human glycophorin A into a [hydropathy plot](/app/hydropathy-plot) with a 19-residue window and one stretch stands out. Positions 99 to 106 average more than 1.6 and peak at 2.67, inside the helix that carries the protein across the red blood cell membrane. The window behind that peak, ITLIIFGVMAGVIGTILLI, holds six isoleucines and not a single charged residue. A second, smaller rise appears at positions 10 to 14. It is not a membrane helix but the signal peptide, which is cut off before the protein reaches the membrane, and the plot has no way to tell the two apart. Both readings come from the same 20 numbers Jack Kyte and Russell Doolittle published in 1982. Reading a hydropathy plot well means knowing what those numbers can separate and what they cannot. ## The Kyte-Doolittle hydropathy scale Kyte and Doolittle gave each amino acid a hydropathy index between +4.5 for isoleucine and −4.5 for arginine. Positive values are hydrophobic, negative values hydrophilic, and the size of the number says how strongly the side chain prefers the protein interior or water. Table 1 lists all 20, ordered from most hydrophobic to most hydrophilic. Table 1. Kyte-Doolittle hydropathy index of the 20 amino acids. Values are from Table 2 of Kyte and Doolittle (1982), where cysteine is listed as cysteine/cystine, and are identical to the Kyte-Doolittle scale in Expasy ProtScale. Amino acid Three-letter code One-letter code Hydropathy index IsoleucineIleI4.5 ValineValV4.2 LeucineLeuL3.8 PhenylalaninePheF2.8 CysteineCysC2.5 MethionineMetM1.9 AlanineAlaA1.8 GlycineGlyG−0.4 ThreonineThrT−0.7 SerineSerS−0.8 TryptophanTrpW−0.9 TyrosineTyrY−1.3 ProlineProP−1.6 HistidineHisH−3.2 Glutamic acidGluE−3.5 GlutamineGlnQ−3.5 Aspartic acidAspD−3.5 AsparagineAsnN−3.5 LysineLysK−3.9 ArginineArgR−4.5 Two features of the table often surprise readers. Tryptophan and tyrosine sit on the hydrophilic side of zero, which Kyte and Doolittle traced to the nitrogen and oxygen atoms in their side chains, and four different residues share exactly −3.5. Neither is a typo; both follow from how the scale was put together. ## How Kyte and Doolittle built the scale Kyte and Doolittle coined the word hydropathy, "strong feeling about water", for the whole range from hydrophilic to hydrophobic. They wanted one number for each side chain's preference between water and the inside of a protein, and they drew on two independent kinds of evidence. The first was the free energy of moving a model compound for each side chain from water into vapor, assembled by Wolfenden and colleagues. They chose vapor over ethanol, used in earlier scales, because ethanol turned out to share many of water's peculiarities. The second was Chothia's 1976 count of how often each residue is fully buried, or more than 95% buried, in 12 globular protein structures. All three columns were rescaled to run from −4.5 to +4.5. The final values then mix arithmetic and judgment. Valine, phenylalanine, threonine, serine and histidine are plain averages of the three columns; cysteine, methionine and isoleucine use the two columns that agreed. Aspartate, glutamate, asparagine and glutamine were judged indistinguishable and given one shared value. Tryptophan received its transfer free energy alone, because the structural data were too uncertain. Other values were set by hand. Glycine got −0.4, the average hydropathy of all sequences in the authors' database, because it showed no clear preference either way. Alanine was lowered, since a single methyl group should not outweigh leucine's four, and arginine was placed at the bottom of the scale. Tyrosine and leucine were raised and lysine lowered, adjustments the authors describe as "the result of personal bias and heated discussion between the authors". They then checked whether any of this mattered. Collapsing the scale into three groups (−3.7, −1.0 and +3.0) left the profiles of chymotrypsinogen and lactate dehydrogenase broadly similar, and raising tyrosine, histidine, proline and tryptophan by 3 units changed them only modestly. They concluded that the second decimal place hardly affects a profile. The values they trusted most, for leucine, isoleucine, valine, phenylalanine, methionine, threonine, serine, lysine, glutamine and asparagine, cover about half of all residues in proteins. They were least sure of cysteine, complicated by disulfide bonds, of proline, aspartate, glutamate and tyrosine, and of arginine, which they suspected should be even more negative. ## How a hydropathy plot is calculated A hydropathy plot is a moving average. For a window of $w$ residues (an odd number) and $k = (w-1)/2$, the value plotted at residue $i$ is the mean hydropathy of the residues from $i-k$ to $i+k$: $$ \bar{h}_i = \frac{1}{w} \sum_{j=i-k}^{i+k} h_j, \qquad k+1 \le i \le L-k $$ Here $h_j$ is the Table 1 value of residue $j$ and $L$ is the sequence length. Each average sits above the middle residue of its window, which is why Kyte and Doolittle used odd window lengths. The first and last $k$ residues have no full window and no point, so a 19-residue window on glycophorin A starts at residue 10 and ends at residue 141. The glycophorin peak shows the arithmetic. The window centred on residue 101 covers residues 92 to 110: six isoleucines (4.5 each), three leucines, two valines, one each of phenylalanine, methionine and alanine, three glycines and two threonines. They sum to 50.7, and $50.7 / 19 = 2.67$. Kyte and Doolittle drew their plots around a midpoint line at the average hydropathy of sequenced proteins, the same −0.4 they gave glycine, rather than at zero. The ProteinIQ Hydropathy plot draws its reference line at zero instead, so a typical soluble protein sits slightly below the line. ## Choosing the window size The window decides what size of feature the plot can show. Kyte and Doolittle compared windows from 3 to 21 residues on proteins of known structure. Windows shorter than 7 were too noisy, long windows missed small consistent features, and windows of 7 to 11 residues best separated a soluble protein's buried segments from its surface. For chymotrypsinogen a window of 9 gave the best match to the crystal structure. Transmembrane helices need a longer window. A segment that crosses the 3 nm hydrocarbon core of a lipid bilayer has to stay hydrophobic for its whole length. An α-helix rises about 1.5 Å per residue, so 19 residues make a helix roughly 28 Å long, close to that thickness. ### Where the 19-residue window and the 1.6 cutoff come from Kyte and Doolittle took the most hydrophobic segment of each of 12 soluble proteins and compared them with nine known membrane-spanning segments: the M13 coat protein, glycophorin and the seven helices of bacteriorhodopsin. The two groups separated most cleanly when the window reached 19 residues. From that comparison they proposed that a 19-residue segment averaging more than +1.6 has a high probability of spanning a membrane. The rule had exceptions from the start. In their own table, the most hydrophobic 19 residues of dogfish lactate dehydrogenase, a soluble enzyme, average 2.26, well above the cutoff. The threshold also belongs to this scale and this window. On another scale, or with a 9-residue window, 1.6 has no meaning. ### How the rule holds up on the human proteome We tested the rule on the reviewed human proteome in UniProt release 2026_03. As true membrane helices we took the 2,191 helical transmembrane segments, in 331 proteins, whose annotation rests on experiments or deposited structures; segments predicted from sequence alone were left out, since those predictions use hydrophobicity themselves. As negatives we took 11,060 proteins with no membrane, transmembrane or signal peptide annotation. ![Line chart of the share of experimentally supported transmembrane helices, signal peptides and soluble human proteins that have a Kyte-Doolittle window above 1.6, for windows of 7 to 21 residues](/images/charts/kyte-doolittle-window-size-cutoff.webp "**Figure 1. What the 1.6 cutoff catches at each window size in the reviewed human proteome.** TM helices: 2,191 helical transmembrane segments with experimental or structural evidence, counted when a window centred inside the segment exceeds 1.6. Signal peptides: 1,607 proteins with a signal peptide and no membrane annotation, counted when a window centred inside the signal peptide exceeds 1.6. Soluble proteins: 11,060 proteins with no transmembrane, intramembrane or signal peptide feature and no membrane keyword, counted when any window exceeds 1.6. UniProt release 2026_03; our calculation.") With a 19-residue window, 1,638 of the 2,191 helices (74.8%) reach 1.6, and 1,330 of the soluble proteins (12.0%) have at least one window above it. Shorten the window to 9 and the rule finds 95.0% of the helices, but 80.4% of soluble proteins now cross the line too. The cutoff only means something with the long window it was calibrated for. About a quarter of real membrane helices still fall short. Many carry polar or charged residues that face other helices or a binding site rather than the lipid, and experiments with the endoplasmic reticulum translocon show that whether a segment inserts depends on where its polar residues sit, not only on its average. A window mean throws position away, which is the price of a method this simple. Our [calculation script](/data/guides/kyte-doolittle-hydropathy/analyze.py) downloads the UniProt data and writes the [counts for every window](/data/guides/kyte-doolittle-hydropathy/window-sweep.csv). ## Reading a hydropathy plot: three proteins The three proteins below cover the cases most plots fall into: a single membrane-spanning helix, a bundle of seven, and a soluble protein with a hydrophobic core. All values use full UniProt sequences and UniProt numbering. ### Glycophorin A: one helix and a signal peptide Glycophorin A (UniProt P02724, 150 residues) has a signal peptide at residues 1 to 19 and one transmembrane helix at 92 to 114. With a 19-residue window, two runs cross 1.6: positions 10 to 14 and positions 99 to 106. Kyte and Doolittle used this protein's helix as their archetype of a membrane-spanning sequence. The run above the cutoff is shorter than the helix because each point averages nine residues on either side. Adding those nine residues to each end of the run, 90 to 115, recovers the annotated helix almost exactly. The same step turns the N-terminal run into residues 1 to 23, which covers the 19-residue signal peptide. Nothing in the plot says which of the two is the membrane anchor; only their position along the chain hints at it. ### Bacteriorhodopsin: seven helices, five peaks Bacteriorhodopsin (UniProt P02945, 262 residues including a 13-residue propeptide) is a light-driven proton pump with seven transmembrane helices. Its 19-residue profile never drops below −0.76, because the loops between helices are short. Only five of the seven helices cross 1.6 (Table 2). Table 2. Highest 19-residue Kyte-Doolittle average inside each bacteriorhodopsin helix. Helix boundaries from UniProt P02945 (release 2026_03), in UniProt numbering, which is 13 higher than the mature-protein numbering common in the bacteriorhodopsin literature. Our calculation. Helix UniProt residues Highest window average Above 1.6 A23 to 421.84Yes B57 to 751.82Yes C93 to 1091.52No D121 to 1401.81Yes E148 to 1671.88Yes F186 to 2041.51No G217 to 2362.12Yes The two misses are the helices that do the chemistry. Helix C carries an arginine and the aspartate UniProt annotates as the primary proton acceptor (Asp98, or Asp85 in mature numbering), and helix F carries an arginine. These polar residues point into the protein, toward the retinal and the proton path, not toward the lipid. Helix G holds the lysine that binds retinal (Lys229, or Lys216 mature) and is still the most hydrophobic of the seven. Counting peaks undercounts helices here for a second reason. The loops at residues 110 to 120 and 205 to 216 never fall below 0.86 and 0.97 in the 19-residue profile, so neighbouring helices merge into one plateau. Kyte and Doolittle hit the same problem in 1982: with a 7-residue window, the last two helices were not resolved, and they placed the turn between them arbitrarily. ### Myoglobin: a soluble protein with a hydrophobic core Human myoglobin (UniProt P02144, 154 residues) is a soluble, oxygen-binding protein built almost entirely of α-helices. Its 19-residue profile peaks at 0.89, far below the cutoff. The peak window, IPVKYLEFISECIIQVLQS, alternates isoleucines, leucines and valines with lysine, glutamate and glutamine: a helix with one face packed against the core and the other in water. With a 9-residue window, myoglobin crosses 1.6 at residues 72 to 73 and 109 to 112 and peaks at 2.27. These short stretches are buried, but they are buried inside the protein, not in a membrane. Short windows show exactly this interior-versus-surface pattern Kyte and Doolittle designed them for, and they are why a soluble protein so often crosses 1.6 at window 9. ### What peaks and valleys tell you A broad peak above 1.6 with a 19-residue window, roughly 20 residues wide once you add the half-window to each side, is a candidate transmembrane helix. A narrower peak at the very start of the chain is often a signal peptide or a signal anchor instead. A plateau that stays above zero over a long stretch, as in bacteriorhodopsin, can hide several helices. Peaks between 0 and 1.6 in the long-window profile usually mark buried segments of a soluble protein or helices with one polar face. Deep valleys mark stretches rich in charged and polar residues, which in a soluble protein lie on the surface; Hopp and Woods used the most hydrophilic six-residue stretches to locate antigenic sites. None of these readings is a prediction on its own. Treat each as a hypothesis to check against a topology predictor or a structure. ## GRAVY: the whole sequence in one number The grand average of hydropathy, GRAVY, is the hydropathy averaged over every residue: $$ \mathrm{GRAVY} = \frac{1}{L} \sum_{j=1}^{L} h_j $$ Kyte and Doolittle introduced it in the same paper. For 84 soluble enzymes, GRAVY did not depend on protein length, and sequenced membrane proteins scored well above them. Yet for membrane proteins known only from their amino acid composition, the spread overlapped so much with soluble proteins that they called the claim that composition alone identifies membrane proteins "tenuous". The three examples show why. Bacteriorhodopsin has a GRAVY of 0.723, myoglobin −0.476, and glycophorin A −0.080, barely different from a soluble protein despite its clear membrane helix. One helix in 150 residues hardly moves the mean. Across the human proteome, the 11,060 soluble proteins in our check have a median GRAVY of −0.505, and 95.7% are below zero. The 331 proteins with experimentally supported membrane helices have a median of 0.170, but 32.6% of them are still negative. A positive GRAVY makes a membrane protein more likely; a negative one rules out very little. The [GRAVY calculator](/app/gravy) returns the value for many sequences at once, and [Protein Parameters](/app/protein-parameters) reports it next to molecular weight, pI and stability indices. ## Where hydropathy plots mislead Signal peptides are the most common false alarm. A signal peptide has a charged N-terminal region, a central hydrophobic core and a polar C-terminal region, and that core looks like a short transmembrane helix to a moving average. In our check, 60.4% of signal peptides in secreted human proteins cross 1.6 with a 19-residue window. Even dedicated predictors confused the two: Phobius, which models both together, cut the share of signal peptides misclassified by TMHMM and SignalP from 26.1% to 3.9%. A hydropathy plot also says nothing about topology, the side of the membrane each loop faces. TMHMM, a hidden Markov model, predicts 97 to 98% of transmembrane helices and their orientation, though its accuracy drops when signal peptides are present. TOPCONS combines several predictors and separates signal peptides from transmembrane regions. DeepTMHMM, built on a [protein language model](/guides/protein-language-models), predicts both α-helical and β-barrel membrane proteins. For the N-terminal peaks, [SignalP 6.0](/guides/what-is-signalp-6) predicts all five known types of signal peptides and their cleavage sites. Run it before calling an N-terminal peak a membrane anchor. Amphipathic helices fall between the categories. A helix that lies along the membrane surface, or packs one face against a protein core, has hydrophobic residues every three or four positions, and a window average dilutes them. Eisenberg and colleagues measured this periodicity as the hydrophobic moment, which is strong at the 3.6-residue period of an α-helix. On their hydrophobic moment plot, single transmembrane anchors are the most hydrophobic with the smallest moment, bundled helices such as bacteriorhodopsin's have larger moments, and surface-seeking helices such as melittin's larger still. ## Kyte-Doolittle compared with other hydrophobicity scales Dozens of hydrophobicity scales exist, and they agree on the extremes and disagree in the middle. Table 3 compares Kyte-Doolittle with four that come up most often. Table 3. Common hydrophobicity scales and what they are for. Ranges for the first three are from Expasy ProtScale. Sources: Kyte and Doolittle (1982); Hopp and Woods (1981); Eisenberg et al. (1984); Wimley and White (1996); Hessa et al. (2005). Scale Based on Positive means Range Typical use Kyte-Doolittle (1982)Water-to-vapor transfer free energies and burial in 12 protein structures, adjusted by handHydrophobic−4.5 (R) to 4.5 (I)Transmembrane helices with a 19-residue window; interior and surface with 7 to 11 Hopp-Woods (1981)Hydrophilicity values tuned to known antigenic sitesHydrophilic−3.4 (W) to 3.0 (R, D, E, K)Antigenic sites, six-residue average Eisenberg consensus (1984)Normalized consensus of published scalesHydrophobic−2.53 (R) to 1.38 (I)Hydrophobic moment and helix classification Wimley-White interfacial (1996)Measured partitioning of small peptides into the interface of zwitterionic phospholipid membranesFree energies, not a normalized indexNot normalizedResidues and peptides at the membrane surface Hessa biological scale (2005)Insertion of designed segments by the endoplasmic reticulum transloconFree energies of insertionNot normalizedWhether a segment inserts as a transmembrane helix Hopp and Woods reversed the sign on purpose: their scale measures hydrophilicity, and they reported that the point of highest six-residue average hydrophilicity lies in or next to an antigenic determinant. Wimley and White measured free energies directly and found aromatic residues especially favoured at the membrane interface, while charged residues and the peptide bond are disfavoured about equally. That interface preference is one reason tryptophan and tyrosine are hard to place on any single scale. Tryptophan shows how far scales can disagree. It scores −0.9 on Kyte-Doolittle, +0.81 on the Eisenberg consensus scale, and −3.4 on Hopp-Woods, which on that inverted scale makes it the least hydrophilic residue of all. Because the ranges differ, a cutoff calibrated on one scale cannot be carried to another. The [Hydrophobicity plot](/app/hydrophobicity-plot) draws the same kind of profile on 24 other scales, including Hopp-Woods and Eisenberg, so you can compare them on one sequence. ## Hydropathy tools on ProteinIQ The [Hydropathy plot](/app/hydropathy-plot) draws the Kyte-Doolittle profile described here, with full windows plotted at the centre residue as in the original paper and Expasy ProtScale, and a window you can set from 5 to 21. Use 19 to look for transmembrane helices and 7 to 11 for surface and interior regions of soluble proteins. Its data table gives the value at every position, which is how the numbers in this guide can be checked. The [Hydrophobicity plot](/app/hydrophobicity-plot) offers 24 other hydrophobicity scales, and [Protein Scale](/app/protein-scale) extends the same moving average to 42 scales, including flexibility, polarity and secondary structure propensity. The [GRAVY calculator](/app/gravy) and [Protein Parameters](/app/protein-parameters) report the whole-sequence average. When a peak needs a structural answer, predict the fold with [ESMFold](/app/esmfold) or [AlphaFold 2](/app/alphafold-2) and look where the hydrophobic segments end up. The [transmembrane protein structure prediction](/use-cases/transmembrane-protein-structure-prediction) workflow combines hydropathy with structure prediction. Start from a clean sequence; [making a FASTA file](/guides/how-to-make-a-fasta-file) covers the format. ## Frequently asked questions ### What are the Kyte-Doolittle hydropathy values? Isoleucine 4.5, valine 4.2, leucine 3.8, phenylalanine 2.8, cysteine 2.5, methionine 1.9, alanine 1.8, glycine −0.4, threonine −0.7, serine −0.8, tryptophan −0.9, tyrosine −1.3, proline −1.6, histidine −3.2, glutamate, glutamine, aspartate and asparagine −3.5, lysine −3.9 and arginine −4.5. Table 1 lists them with their codes. ### What window size should I use for a hydropathy plot? Use 19 residues to look for transmembrane helices, the window Kyte and Doolittle calibrated with the 1.6 cutoff. Use 7 to 11 residues to see which parts of a soluble protein are buried or on the surface; 9 worked best on chymotrypsinogen in the original paper. ### What does a hydropathy value above 1.6 mean? With a 19-residue window, a segment averaging above 1.6 is hydrophobic enough to be a membrane-spanning helix. In the reviewed human proteome, that rule catches about three quarters of experimentally supported membrane helices, flags 12% of proteins with no membrane annotation, and fires on most signal peptides. Confirm candidates with a topology predictor. ### Is a positive hydropathy value hydrophobic? Yes. On the Kyte-Doolittle scale positive values are hydrophobic and negative values hydrophilic. Hopp-Woods uses the opposite sign, so check which scale a plot uses before reading its peaks. ### Can a hydropathy plot predict membrane protein topology? No. It shows where hydrophobic segments are, not which side of the membrane each loop faces, and it can miss helices with polar residues or merge neighbouring helices. Topology predictors such as TMHMM, TOPCONS and DeepTMHMM are built for that question. ### What is a good GRAVY score? There is no good or bad value. Most soluble proteins are negative: in our human proteome check, the median soluble protein has a GRAVY of −0.505. Positive values are more typical of membrane proteins, but a third of human proteins with experimentally supported membrane helices are still below zero.