Hi Peter, what does "score: a likelihood based score combining all info" exactly mean? What is the "good value" (and why), what is "so-so value", and what is the "bad value" (and why)? Could you please give a reference to where this score is defined and its use is evaluated? I asked myself all these questions when I read this phrase in your reply below and, sorry for my ignorance, I failed to find the answers. Thanks! Pavel. On 4/20/09 12:06 PM, Peter Zwart wrote:

2009/4/20 :

...

Hi,

I tried running phenix xtriage with a scaled data in P4322 space group. I got the following output: ----------------------------------------------------------------------------------- | space group | n absent | <Z>_absent | _absent | +++ | --- | score | ----------------------------------------------------------------------------------- | P 4 2 2 | 0 | 0.00 | 0.00 | 0 | 5 | 0.000e+00 | | P 41 2 2 | 0 | 0.00 | 0.00 | 0 | 5 | 0.000e+00 | | P 42 2 2 | 0 | 0.00 | 0.00 | 0 | 5 | 0.000e+00 | | P 43 2 2 | 0 | 0.00 | 0.00 | 0 | 5 | 0.000e+00 | | P 4 21 2 | 8 | 1.71 | 8.07 | 6 | 3 | 4.817e+00 | | P 41 21 2 | 8 | 1.71 | 8.07 | 6 | 3 | 4.817e+00 | | P 42 21 2 | 8 | 1.71 | 8.07 | 6 | 3 | 4.817e+00 | | P 43 21 2 | 8 | 1.71 | 8.07 | 6 | 3 | 4.817e+00 |

Can anyone explain what these numbers mean? It seems all enantiomorphic space groups have the same score (which is the last column)

n absent: number of absent intensities according to spacegroup

Z absent: mean normalized intensity for absent reflections Z/sigZ: 'signal to noise' of absent reflections +++ : reflections that seem present that should be absent ---- : reflections that seem absent but should be present score: a likelihood based score combining all info

If you want, you can send me your sca file and I can have look to make sure that nothing is messed up.

Cheers

Peter

...

The documentation in phenix does not explain this in detail. I scaled the data in HKL2000 and there was no systematic absences with P43222. Thanks, Shya

_______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb

Hi Pavel, Normalized intensity is normalized as usual (includes epsilon weights): <Z> = 1. When a reflection is classified as absent due to the space group, its ideal value should be zero. Any deviation from zero would be due to experimental error. If a reflection is not absent, one expects that this intensity would be drawn from a Wilson like distribution (either centric or acentric), depending on the symmetry. The 'error' free value of the intensity responsible for the observation, is thus either 0 (when absent) or drawn from a Wilson distribution. The 'score' for a single observation (depending on the space group), is then (for an observed intensity) - int_0^\infty log[p(z | z_obs, sigma_z)]*p(z| wilson) d z with p(z | z_obs, sigma_z) conveniently designated as a Gaussian The best score is used as a base line, and the best scapcegroup will always be zero. Strictly speaking, a score of 4.8 for P43212 indicates that P422 (with score 0) is about 120 (exp[score]) times more likely. Realistically, the order is more important and can be used to prioritize MR or HA searches if need be. Also, when no absence information is present (missing data) all space groups will have the same score. It provides a hassle free way to rank possible space group on the basis of the associated implications for intensity statistics. Below you will find a table with scores for a single reflection with observed normalized intensity in the first column (Z), its score when it is acentric, centric or absent. In all cases, sigma was 0.5. Lower score indicate a more likely choice. In this case, an observed value of 0.5 indicates that the reflection is more likely to be centric then absent. If z would be 0.4, it would be more likely that it would be an absent reflection. Z acen cen absent z/sigz 0.00 1.00 0.67 0.23 0.00 0.05 0.94 0.64 0.23 0.10 0.10 0.89 0.61 0.25 0.20 0.15 0.84 0.59 0.27 0.30 0.20 0.80 0.58 0.31 0.40 0.25 0.77 0.57 0.35 0.50 0.30 0.74 0.58 0.41 0.60 0.35 0.72 0.58 0.47 0.70 0.40 0.70 0.60 0.55 0.80 0.45 0.69 0.62 0.63 0.90 0.50 0.68 0.65 0.73 1.00 0.55 0.69 0.68 0.83 1.10 0.60 0.69 0.71 0.95 1.20 0.65 0.70 0.75 1.07 1.30 0.70 0.72 0.79 1.21 1.40 0.75 0.73 0.84 1.35 1.50 0.80 0.76 0.89 1.51 1.60 0.85 0.79 0.94 1.67 1.70 0.90 0.82 1.00 1.85 1.80 0.95 0.85 1.06 2.03 1.90 1.00 0.89 1.12 2.23 2.00 The total score for a space group is based on the different classifications for the miller indices. I have found this scheme to be more reliable then plain i/sigi considerations, as it takes into account differences between centric vs acentric. Furthermore, no special precautions need to be taken for missing data, nor do I need to worry about completion issues when usnig FFT based approaches.

"bad value" (and why)? Could you please give a reference to where this score is defined and its use is evaluated?

I asked myself all these questions when I read this phrase in your reply below and, sorry for my ignorance, I failed to find the answers.

Not a surprise, this still needs to be published. HTH Peter -- ----------------------------------------------------------------- P.H. Zwart Beamline Scientist Berkeley Center for Structural Biology Lawrence Berkeley National Laboratories 1 Cyclotron Road, Berkeley, CA-94703, USA Cell: 510 289 9246 BCSB: http://bcsb.als.lbl.gov PHENIX: http://www.phenix-online.org CCTBX: http://cctbx.sf.net -----------------------------------------------------------------