2,684
edits
Xds nonisomorphism: Difference between revisions
expand
No edit summary |
(expand) |
||
| Line 20: | Line 20: | ||
== Analysis and interpretation == | == Analysis and interpretation == | ||
Angles (calculated as the inverse cosine of the ratio) are expressed in degrees. Less than 10° may be considered good isomorphism, 90° means highly non-isomorphous (i.e. completely unrelated) datasets. However, the interpretation of the magnitude of an angle depends on the resolution. To | Angles (calculated as the inverse cosine of the ratio) are expressed in degrees. Less than 10° may be considered good isomorphism, 90° means highly non-isomorphous (i.e. completely unrelated) datasets. However, as seen in actual tables, the numerical value (and the interpretation of the magnitude of an angle) depends on the resolution. But there is another interpretation of the ratio (column 6) - not as cos(phi) but as a correlation itself. To make sense of this interpretation, the program uses a formula (McCoy et al. (2017) PNAS 114, 3637-3641 equation 1) that relates coordinate difference to correlation (column 8 of output). This coordinate RMSD value should be independent of resolution. If it is (which is sometimes seen in pairwise comparisons of data sets) then this is an indication that some other systematic difference, that cannot be interpreted as coordinate difference, exists between data sets. Candidates are many kinds of sources of systematic error, e.g. errors in data processing, twinning, overloads, vibrations ... | ||
After the analysis, the program produces a 3D representation of the arrangement of data sets such that their distances in 3D try to reproduce the angles (please note that this is | After the analysis, the program produces a 3D representation of the arrangement of data sets such that their distances in 3D try to reproduce the angles (please note that this representation is completely different from that of [[xscale_isocluster]]!). | ||