Hydropathy Scales: Kyte-Doolittle, Hopp-Woods & Eisenberg

Which scale answers which question

The sign convention is the trap. Kyte-Doolittle and Eisenberg both run positive for hydrophobic residues, so arginine sits at the bottom of both. Hopp-Woods measures the opposite property, hydrophilicity, so arginine is its highest value at +3.0. Reading a Hopp-Woods profile as though it were Kyte-Doolittle inverts every conclusion.

Window size and the transmembrane threshold

Both numbers people quote — a 19-residue window and a +1.6 cut-off — are in the original 1982 paper. They are not a later ExPASy convention, which is what several pages currently say. Kyte & Doolittle scanned every span from 9 to 21 residues over 12 soluble proteins and 9 known membrane-spanning segments, and wrote (p. 119):

“The discrimination between the segments from the soluble proteins as a group and those from the membrane-spanning sequences was most unequivocal when the span was lengthened to 19 residues… when the hydropathy of a given 19-residue segment averages greater than +1.6 there is a high probability that it will be one of the sequences in a membrane-bound protein that spans the membrane.”

“High probability”, not “is”. The threshold is an observation drawn from one small table, never a validated decision rule — there is no sensitivity, specificity or held-out set behind it. And the table it came from contains a miss and a false alarm at that very cut-off:

So “most unequivocal at 19” means least bad, not clean. Use the plot to see where the hydrophobic stretches are; use a topology predictor to decide whether they cross a membrane and which way they face.

Three windows, three questions, three authorities

• 19 — does this segment cross a membrane? Threshold +1.6 on the window average, from the passage above. Nineteen residues is about the length of an α-helix spanning a bilayer.
• 9 — is this residue buried or exposed?Also from the same paper, and routinely misattributed. Comparing profiles of chymotrypsinogen against its crystal structure, “the best agreement between strongly hydrophobic segments and interior regions and strongly hydrophilic segments and the exterior was obtained with a setting of nine residues” (p. 113). This is why 9 is the default almost everywhere, including SeqBench’s own 9-residue window — and why a 9-residue profile is not a transmembrane screen.
• 6 — where is a surface epitope?A different question, a different scale and a different paper: Hopp & Woods report that prediction success depended on the averaging group length and that hexapeptide averages gave the optimal result. Do not attribute this window to Kyte & Doolittle.

Two details the original states that implementations get wrong. Spans are odd “so that a given sum could be plotted above the middle residue of the segment” (p. 107). And the midpoint line is not zero — it is the grand average of the amino-acid compositions of sequenced proteins, which the paper puts at −0.4, while most plotters draw the midline at 0.

Hopp-Woods is not Kyte-Doolittle with the sign flipped

The two scales are strongly anti-correlated — Pearson r = -0.70 across the twenty residues — which is why swapping them usually produces a profile that looks upside-down but plausible. They are not, though, related by a sign change: 4 residues (His, Thr, Trp, Tyr) are negative on both scales, which no inversion can produce. Tryptophan reads mildly hydrophilic on Kyte-Doolittle at -0.9 and strongly un-hydrophilic on Hopp-Woods at -3.4. Eisenberg, by contrast, does track Kyte-Doolittle closely (r = +0.88) on a compressed range.

Sources

1. 1
   A simple method for displaying the hydropathic character of a protein

   Kyte J, Doolittle RF · J Mol Biol 157:105-132 · 1982

   The Kyte-Doolittle column — all 20 values, checked residue by residue against ExPASy ProtScale and Biopython's ProtParamData, two independent transcriptions that both cite this paper. Also the 19-residue window, the +1.6 cut-off, the 9-residue window for globular interiors, and Table 4: all four were read off the typeset pages (107, 113, 119, 120) of the author copy, not taken from a secondary source.
2. 2
   Prediction of protein antigenic determinants from amino acid sequences

   Hopp TP, Woods KR · Proc Natl Acad Sci U S A 78:3824-3828 · 1981

   The Hopp-Woods column — all 20 hydrophilicity values, and the 6-residue averaging window: its abstract states that prediction success depended on averaging group length and that hexapeptide averages gave optimal results.
3. 3
   Analysis of membrane and surface protein sequences with the hydrophobic moment plot

   Eisenberg D, Schwarz E, Komaromy M, Wall R · J Mol Biol 179:125-142 · 1984

   The Eisenberg column — the normalized consensus hydrophobicity scale, all 20 values. This is the paper the numbers correspond to; Eisenberg published more than one scale, and the consensus scale is the one ExPASy and Biopython serve under this name.
4. 4
   ExPASy ProtScale — Hydropathicity (Kyte & Doolittle)

   SIB Swiss Institute of Bioinformatics

   Not a source for the numbers but the cross-check on them: ProtScale's own per-scale pages were compared value by value against this table before it was published, and all sixty matched.

Related tools and references