The shape_screen and shape_screen_gpu Commands

The command syntax for a shape search on a CPU host is as follows:

$SCHRODINGER/shape_screen -JOB jobname {CHECKPOINT|-RESTART path|-NO_CHECKPOINT} [program-options] [job-options] 

The command syntax for a shape search on a GPU host is:

$SCHRODINGER/shape_screen_gpu task [program-options] [job-options]

The main job options are described in Running Jobs From the Command Line. The program options, other job options, and the required arguments are described in shape_screen Command Help and shape_screen_gpu Command Help, and can be listed by running the command with the -h option. You can run shape screening from Maestro—see Shape Screening Panel for details.

CPU screens

Two screening algorithms are available for CPU screens, which are run by different programs. The new algorithm is much faster, and is used by default (fast_shape). You can use the previous algorithm by adding the -classic option to the command. The -classic option is required if you are screening a database in the older Phase format (_phasedb). There are slight differences in the similarities and alignments between the two methods. Some of the options available with the classic algorithm are not available with the new, fast algorithm (e.g., -report, -table, -force). If you select any of these options, the classic algorithm is used.

When you run a job with the -classic option, you must specify whether you want to create checkpoint files or not. If you do not create checkpoint files, the job cannot be restarted. To restart a job, you can use the syntax:

$SCHRODINGER/shape_screen -JOB jobname -RESTART path [-osub dbSubOut] [job-options] 

The new, fast algorithm does not use this mechanism; instead it automatically restarts failed subjobs.

The aligned structures are written to the file jobname_align.maegz. For each molecule, the alignments for a given query are stored consecutively, ordered by similarity. If sorting is requested, the blocks of alignments for each query are ordered by similarity, and the molecules are ordered by the maximum similarity for any query or alignment for that molecule. If sorting is not requested, the blocks of alignments for each query are in the order that the queries appear in the query input, and the molecules are in the same order as the molecules in the input.

To set up the file required for a SMARTS prealignment, you can display the query in Maestro, label the atoms with the atom numbers, then use the Find toolbar to find the desired SMARTS patterns. You can then copy the string and the atom numbers to the file. The atom numbers must be in the same order as the atoms in the SMARTS string.

GPU screens

Shape screening on a GPU requires a shape data file that contains the shapes for the ligands to be screened, in a format suitable for GPU use. This shape data file can be generated with the generate task. The options for the generate task specify all the information required for setting up the ligand conformers and creating shapes, much like the generation of a Phase database. You can split the shape data file up into chunks by using the -divide option. This allows you to run the screen on multiple GPUs.

Once the shape data file is created, you can perform the screening, with the run task. The classic algorithm (see above) is used for GPU screens. When you run the job, you can specify multiple shape data files, and the driver creates a subjob for each file. If the queueing system is set up appropriately, you can run jobs that use multiple GPUs on multiple nodes. The shape data files must be in a location where they are accessible from the nodes.

For full details of the command options, run the command

$SCHRODINGER/shape_screen_gpu task -h