PSD

What it does. This tool predicts a protein's secondary structure — the proportion of its shape made up of Helix, Sheet, Turn, and Other regions — from an uploaded spectrum. It works by comparing your spectrum against a library of reference spectra with known, verified structures, and finding the closest matches.

1. Prepare your file. Your data should be a plain text file (.txt, .csv, .tsv, or .dat) with one data point per line, containing two numbers per line:

• The wavenumber (in cm⁻¹)
• The absorbance value at that wavenumber

Separate the two numbers with either a tab or a comma. For example:

1700	45.2
1699	46.1
1698	47.0
...

2. Check your wavenumber range. Your spectrum must include the 1500–1700 cm⁻¹ range (the "Amide I/II" region). This is the window shared by every reference spectrum in the library, and is what the comparison is based on. Data outside this range is fine to include, but won't be used. If your file doesn't cover this range, the tool will show an error rather than a result.

3. Upload and run. Drop your file into the box below (or click to browse), then click "Run prediction." Results usually appear within a few seconds.

4. Reading your results. You'll get back four percentages — Helix, Sheet, Turn, and Other — which together add up to 100% of the protein's structure. Below that, "Nearest reference spectra" shows which known spectra your upload was compared against, and how closely each one matched (as a percentage similarity). A very high similarity (99%+) to one particular reference usually means a strong, confident match; lower similarities across several references mean the prediction is more of a blend and worth treating with a bit more caution.

A note on accuracy. Predictions are generated using a spectral-correlation method against a reference library of known spectra, not a wet-lab measurement — treat results as a helpful estimate rather than a definitive structural determination.