Chou-Fasman

What is the Chou-Fasman method?

The Chou-Fasman method predicts protein secondary structure from amino acid sequence alone. Published by Peter Chou and Gerald Fasman in 1974, it was the first widely used algorithm to show that local amino acid composition carries enough information to predict helices, sheets, and turns.

The approach is statistical rather than physical. Chou and Fasman counted how often each amino acid appeared in known crystal structures within helices, sheets, or turns, then converted those frequencies into propensity scores. A propensity above 100 means that amino acid favors the structure; below 100 means it disfavors it. These 60 numbers (20 amino acids times 3 structure types) are the entire model.

Accuracy sits around 60-65%, well below modern deep learning methods like ProstT5 or ESMFold. The method remains useful as a teaching tool and as a fast, interpretable baseline. Because it uses no MSA, no neural network, and no GPU, predictions return in seconds.

How to use Chou-Fasman online

ProteinIQ runs the canonical Chou-Fasman algorithm (1978 parameter set) on its servers. Paste one or more sequences and results are returned within seconds.

Inputs

Output columns

Results can be downloaded as CSV, JSON, or copied directly.

How does Chou-Fasman work?

The algorithm runs in four steps, applied sequentially to each sequence.

1. Nucleation

The method scans for local stretches rich in structure-forming residues:

• Helix nucleation: At least 4 of any 6 consecutive residues must have $P_\alpha > 100$.
• Sheet nucleation: At least 3 of any 5 consecutive residues must have $P_\beta > 100$.

These thresholds reflect the cooperative nature of secondary structure formation. A single strong former surrounded by breakers won't nucleate.

2. Extension

Each nucleated region extends outward one residue at a time in both directions. Extension continues as long as the average propensity of 4 consecutive residues at the boundary stays above 100. Once this running average drops below the threshold, the region terminates.

3. Turn prediction

Turns are predicted for each tetrapeptide (4-residue window) that satisfies all three conditions:

• The product of positional turn frequencies exceeds $7.5 \times 10^{-5}$. These frequencies capture the preference of each amino acid for specific positions within a turn.
• The average turn propensity $P_t$ of the four residues exceeds 100.
• The sum of $P_t$ values exceeds both the sum of $P_\alpha$ and the sum of $P_\beta$ for the same four residues. This ensures turns are only predicted where turn tendency genuinely dominates.

4. Conflict resolution and cleanup

When predicted helix and sheet regions overlap, the algorithm compares the total helix propensity ($\sum P_\alpha$) against total sheet propensity ($\sum P_\beta$) across the overlap. The higher sum wins. After resolution, any helix or sheet region shorter than 5 residues is removed as unreliable.

Positions not assigned to helix, sheet, or turn default to random coil (C).

Understanding the results

The Structure string is the primary output. Each character maps directly to a position in the input sequence:

• H (helix): The residue is in a predicted alpha-helix region
• E (sheet): The residue is in a predicted beta-sheet region
• T (turn): The residue is at a predicted turn
• C (coil): No regular structure predicted

Typical globular proteins show 30-40% helix and 15-25% sheet. If a prediction shows 0% for both, the sequence may be intrinsically disordered, or too short for meaningful nucleation (sequences under ~20 residues often produce all-coil predictions).

The percentage and region count columns are useful for comparing across sequences. A protein with 35% helix in 4 regions has a different architecture than one with 35% helix in 1 long region.

When to use Chou-Fasman vs alternatives

Chou-Fasman is a sequence-only, statistics-only method. Several alternatives exist depending on what matters most:

• ProstT5 uses a protein language model and achieves over 80% accuracy for 3-state (H/E/C) prediction. For any serious research application, ProstT5 is the better choice.
• DSSP assigns secondary structure from a solved 3D structure (PDB file), not from sequence. If a crystal or cryo-EM structure exists, DSSP gives ground truth rather than a prediction.
• ESMFold and AlphaFold 2 predict full 3D structures, from which secondary structure can be derived. These are slower but far more informative.

Chou-Fasman's niche is speed and interpretability. Every prediction can be traced back to a specific propensity value and no black-box model is involved. This makes it well-suited for teaching the fundamentals of structure prediction, for quickly screening large sequence sets, and for situations where understanding why a prediction was made matters more than accuracy.