TM-score and RMSD

Purpose

The TM-score [1] and RMSD are two distinct metrics used to evaluate the structural similarity between protein models or conformations. RMSD measures the average distance between the backbone atoms of superimposed structures; a lower RMSD indicates greater similarity, but it is highly sensitive to local deviations and misaligned regions, making it less reliable for assessing overall fold similarity, especially when large conformational changes or flexible regions are present. In contrast, the TM-score is designed to be sensitive to the global topology of the protein. It is less affected by local errors or misalignments, providing a more robust measure of overall structural resemblance. It is often preferred for comparing structures with significant differences or for assessing the quality of protein models.

The results are presented as the average values over the trajectory. Evolution of the metrics over time is additionally plotted.

Description

The implementation works as follows:

  • The trajectory is loaded as an mdtraj.Trajectory object (see mdtraj).

  • Carbon alpha atoms are extracted from the trajectory

  • Both RMSD and TM-score are computed on the extracted frames using the compute_tm_scores function of tmtools

Interpretation

The TM-score ranges from 0 to 1, where 1 indicates a perfect match and scores above 0.5 generally suggest a similar fold. Anything below 0.5 is considered a poor match.

A RMSD closer to 0 indicates a better match to the reference structure.

References