Appendix C: TNT scripts used by BUSTER

Content

• Introduction
• tntlongll
• tntshortll
• tntsfcalc
• Changing the command inputs to the TNT scripts

Introduction

1. tntlongll carries out the operations previously performed by the long loop in the standard TNT script;

2. tntshortll carries out the operations previously performed by the short loop in the standard TNT script;

   Both of these scripts are applied to the sublist of reflexions defining the working set, i.e those reflexions for which FreeR_flag is different from the user's defined value flagging the test set.

3. tntsfcalc calculates the structure factors from the fragment.

A brief description of these scripts will now be given. Any necessary changes would probably be in the command input sections to the TNT modules rfactor, geometry, ncs and shift.

tntlongll

1. If hard NCS is requested, the input model is folded into a single copy (the protomer) using the NCS operators given by the user.

2. Gradient and curvatures of the Log-Likelihood Gain (LLG) or of the Least-Squares residual are computed in the form of "X-ray residual" suitable for ML or LS refinement according to user's choice. For ML refinement, this is done in 3 steps:

   a)	compute gradients for the working-set reflexions by running the rfactor module (this is using the AGARWAL command with the LLG gradient information llgrad.pak produced by BUSTER and the expectation values of the model amplitudes expect.pak according to the current hypothesis); input commands are given in-line and the output file is agarwal.dat.
   b)	compute gradients and curvatures for the working-set reflexions by running the rfactor module (this is using the CURVATURE command with the LLG gradient information llcurv.pak produced by BUSTER); input commands are given in-line and the output file is called rfactor.dat (which after the following step is renamed to rfactor.old).
   c)	merge the gradients computed at step a) and the curvatures computed at step for b) into a single file rfactor.dat of gradients and curvatures of the LLG.

   When least-square (LS) refinement is selected, the gradients and curvatures can be produced in a single call to the rfactor module.

   If hard NCS is used, the GRADIENT and CURVATURE cards are folded by the ncs module to produce a single set of gradients are curvatures for the minimiser.

3. Calculation of gradients and curvatures of the GEOM stereochemical restraint criterion. The commands are given in-line and the output is geometry.dat. When hard NCS is requested, the GRADIENT and CURVATURE cards are folded by the ncs module to produce a unique set of gradients and curvature for the minimiser. tntlongll at this stage also screens the worst agreements between model and target values of stereochemical "observations" When soft NCS is selected, the screening is extended to NCS equivalence between NCS-related copies.

4. If soft NCS is requested, the program ncs is used to compute gradients and curvatures of the NCS residual between various copies of the protomer.

5. Calculation of parameter shifts. The gradients and curvatures of the X-ray residual (LLG or LS), of GEOM and opionally of NCS are read, and a shift direction is computed and applied. If the NCS is constrained, the shifted protomer is expanded by NCS operators.

tntshortll

1. The X-ray residual is computed only for LS refinement.
2. The stereochemistry is evaluated, and
3. The NCS residual is also evaluated if soft NCS was chosen.
4. The coordinates are shifted again, and expanded by NCS operators if necessary.

tntsfcalc

Changing the command inputs to the TNT scripts

• You might want to suppress the VERBOSE option in the tntlongll and tntshortll scripts, if you find the TNT logfile tntlongll.html too unwieldy.

• On the contrary, if additional information is wanted in the main output, the script can be modified to monitor specific information, for example the agreement between NCS-related copies when NCS is restrained.