Help

This tool predicts the stability of the collagen triple helix from amino acid sequence, using an algorithm derived from experimental host-guest peptide thermal stability data. It offers two related calculations, described below.

Predicting T_m for short peptides

This mode applies to a peptide with a repeating (Gly-X-Y)_n sequence and predicts its melting temperature, T_m, under standard conditions (c = 1 mg/ml; PBS, pH 7; average heating rate 0.1°/min). It is intended for peptides of up to 30 tripeptide units - longer sequences should use the Stability Profile tool instead.

If either terminus of the peptide is chemically blocked, indicate this using the corresponding checkbox, since blocking increases stability and changes the predicted T_m.

The base T_m for a fully stable (Gly-Pro-Hyp)_n peptide of \(n\) triplets is:

\[ T_m^{(0)} = y_0 + a_1 \, e^{-n/t_1} \]

This is then adjusted using the measured stability, \(S_i\), of each internal triplet (excluding the first and last, which are not fully constrained by neighbors):

\[ T_m = T_m^{(0)} - \sum_{i=2}^{n-1} (47.3 - S_i) \]

Generating a stability profile for longer sequences

Collagen molecules are long and consist of multiple domains, so a single T_m value cannot describe them. Instead, this mode computes a relative stability value for each tripeptide position along the chain - a measure of how stabilizing or destabilizing that local region is, compared to a fully stable (Gly-Pro-Hyp)_n reference, based on the same host-guest peptide data used for T_m prediction. These values reflect relative thermal stability along the chain and should not be interpreted as absolute melting temperatures.

Because triplet-by-triplet values are noisy, they are smoothed across a window of neighboring triplets before plotting. Two settings control this:

Averaging window - how many neighboring triplets contribute to each point on the curve. A small window (e.g. 1-5) preserves sharp local detail; a larger window (e.g. 15-25) shows broader trends and suppresses noise.

Smoothing function - how those neighboring values are combined:

• Rectangular - an unweighted moving average over the window \(w\): \[ \bar{S}_i = \frac{1}{w} \sum_{j=i-\frac{w-1}{2}}^{i+\frac{w-1}{2}} S_j \] (near the ends of the sequence, out-of-range terms are dropped from both the sum and the denominator, rather than treated as zero.)
• Gaussian - a normalized, distance-weighted average, using a Gaussian kernel of width \(w\): \[ \bar{S}_i = \frac{1}{\sigma\sqrt{2\pi}} \sum_{j} S_j \, e^{-(i-j)^2 / w^2}, \quad \sigma = \frac{w}{\sqrt{2}} \]
• Cumulative Gaussian - the same Gaussian kernel applied to each triplet's deviation from the sequence's overall mean stability, \(\bar{S}\), without normalizing by the total kernel weight, then added back onto the mean: \[ \bar{S}_i = \bar{S} + \sum_{j} (S_j - \bar{S}) \, e^{-(i-j)^2 / w^2} \] Because this sum isn't divided by the total kernel weight, its magnitude is more sensitive to the choice of window size than the other two methods - useful for exaggerating local peaks and dips, but less directly comparable across different window settings.

The tool also supports collagen heterotrimers: enter three chains of equal length and leave the homotrimer checkbox unselected, and the three chains' stabilities are averaged at each position before smoothing.

Sequence input format

Enter sequences using standard one-letter amino acid codes, in any format - for example, "GPOGARGMOGPAGPOGAR" or "GPO-GAR-GRE-GTA-GTT-GAA-GPO" are both accepted. Any characters other than standard amino acid codes, including numbers and punctuation, are ignored. If a sequence contains a break in the G-X-Y repeating pattern, everything after the break is disregarded - check the resulting sequence length to confirm it was parsed as intended.

Use "O" to denote hydroxyproline (Hyp) directly, or enter unmodified proline and select the corresponding checkbox to convert all Y-position prolines to hydroxyproline automatically, reflecting typical post-translational modification.

Accuracy

While every effort has been made to ensure the accuracy of the information and predictions provided on this site, we offer no guarantees and strongly encourage independent verification of all sequences and results before relying on them.

Citing this tool

The underlying algorithm is described in Persikov, A., Ramshaw, J.A. and Brodsky, B., J. Biol. Chem. (2005), Prediction of collagen stability from amino acid sequence. Please cite this reference in any publication that includes predictions generated by this website. See the References page for the full citation and related patent.