For cryo-EM models

It’s important to note that sometimes models are only built for parts of a 3DEM map, especially if some regions are less well-resolved. A model fitted or built into a map can be scored on how well it matches the map density in several ways.

Fourier Shell Correlation (FSC)

This is the primary and most objective way to estimate for single-particle and subtomogram averaging methods. It quantifies the correlation between two independently reconstructed half-maps (each reconstructed using half the data) at different levels of detail (spatial frequency). A higher correlation (closer to 1) at higher resolutions (i.e., smaller numbers in Ångströms, representing more detail) indicates a better quality and more reliable data.

The shape of the curve depends on the imposed symmetry, mask and whether or not the two 3D reconstructions used were processed from a common reference. An FSC curve typically starts high (near 1.0) at low spatial frequencies (low resolution, large features) and gradually drops towards zero at high spatial frequencies (high resolution, fine details). The point where the curve crosses a specific threshold (The community agreed value is 0.143, but 0.5 can be used for a more conservative estimate) is typically taken as the map’s resolution.

An idealised FSC curve tends to be thought to mimic the second half of a normal distribution. Sharp drops or jagged sections can suggest problems such as overfitting or artefacts introduced by masking.

Q-score

This is a very useful metric specifically designed for cryo-EM maps, providing a local (per-atom) measure of how well an individual atom or even a single amino acid residue is resolved within the 3DEM map. It primarily indicates the atomic resolvability of the map at that location. Q-scores can be calculated for protein and nucleic acid atoms, waters, ions, etc., using models fitted to or derived from cryo-EM maps.

If a model is misaligned or an atom is misplaced, it will cause the Q-score to be calculated in the wrong location, thus indicating a problem with the model building. A higher Q-score (closer to 1) indicates a higher confidence in the atomic resolvability in that specific region, whilst closer to 0 indicates low similarity. If the atom is not well resolved in the map, then a negative Q-score value may be reported. Therefore, Q-score values in the reports will typically range from -1 to +1.

Users can also investigate the dataset by accessing the raw images stored at EMPIAR, the Electron Microscopy Public Image Archive.