Shape based molecular screening using GPU

This tutorial will show how to use the Schrödinger Maestro Shape tool to quickly screen the Molport molecular database with 16 million structures (160 million conformations) on a GPU in Puhti.

Preparing the Molport database for GPU screening takes two weeks, but since the prepared shape file is already available on Puhti the actual screening of your molecule takes only 5-20 minutes on a GPU. Read more in a blog on how sharing enables fast screening .

HOW-TO

Install Maestro on your local computer. Detailed instructions can be found on our Maestro page.
Draw / import the molecule you want to find similar molecules of
Run the Maestro LigPrep tool on it
Export the prepared molecule in the native Maestro format (mae), e.g. as target.mae
Copy the file to Puhti in your scratch area (Tips for copying files)
Copy (and edit if necessary) the following script as gpushape.bash (don't copy the 139GB shape-file) to the same folder
- This script will request 60 minutes of time on one GPU, use the shape file from /appl/data/bio/molport-shape/MP_SHAPE.bin, use target.mae as the search molecule, and put the results in shape_test-out.maegz
```
$SCHRODINGER/shape_screen_gpu run -shape_data_treatment remote \
      -shape target.mae -HOST quick-gpu -JOBNAME shape_test \
      -screen /appl/data/bio/molport-shape/MP_SHAPE.bin
```
Initialize the default Maestro installation with
```
module load maestro
```
...and submit the job (note, without sbatch!)
```
bash gpushape.bash
```
Copy the results to your local computer for analysis