Skip to content

Structure alignment

Superpose two structures and score how similar their folds are, with TM-score.

First structure

Or drop a file here.

Second structure

Or drop a file here.

Load two structures to align them. Neither file is uploaded.

FormulaTM-score = (1/L) Σ 1/(1 + (dᵢ/d₀)²); d₀ = 1.24·∛(L−15) − 1.8
ModelTM-align: iterative dynamic programming over a distance-weighted score matrix, with closed-form superposition

TM-score is asymmetric — it depends which length you divide by — so both are shown rather than one being chosen for you. Above about 0.5 two structures share a fold; below about 0.3 they are no more alike than two proteins picked at random. Only alpha carbons are used. This is an independent implementation of the published method and will not agree with the reference program to the third decimal.

When to use this

Use this to superpose two structures and measure how similar they are, without needing their sequences to match. Reach for TM-score when the question is whether two proteins share a fold, and RMSD when they are the same protein in two states.

Worked example

Comparing the alpha and beta chains of haemoglobin.

First structure
4HHB chain A, 141 residues
Second structure
4HHB chain B, 146 residues

Result

TM-score 0.90, RMSD 1.41 Å over 139 residues

The globin fold, found from geometry alone at only 43% sequence identity — which is the point of a sequence-independent alignment.

What people get wrong

  • Judging fold similarity by RMSD. It grows with length and one flexible loop drags it as hard as a wrong topology; 4 Å means different things across 60 and 600 residues.
  • Quoting a TM-score without saying which length normalised it. The measure is asymmetric, and a small domain inside a large protein scores high one way and low the other.
  • Reading a single number and stopping. The per-residue plot is where a rigid core with one moving hinge separates from two structures differing everywhere.

Questions

+What TM-score counts as the same fold?

Above about 0.5 two structures share a fold; below about 0.3 they are no more alike than two proteins picked at random.

+Are my files uploaded anywhere?

No. Both are read locally with FileReader and never leave your machine, which for an unreleased structure is the difference between a usable tool and one your institution will not allow.

+Does it accept mmCIF?

Yes, and PDB. The RCSB has served mmCIF by default since 2019, and large structures cannot be represented in the older fixed-column format at all.

Related tools

Science last reviewed .