[ RIDE Setup ]

RIDE is a fast 3D molecular similarity search method based on Atomic Property Fields, developed at MolSoft. RIDE searches databases of compound conformers for molecules that are isosteric to the query, i.e. have similar 3D configurations and distributions of atomic properties.

GPU-based implementation is capable of searching ~0.5mln conformers/second on a single GPU card and perform 3D virtual screens of millions of compounds with the level of interactivity comparable to 2D searches. VLS benchmarking on DUDe indicates that RIDE searches produce enrichments on par with much slower standard flexible APF VLS.

RIDE applications include:

• Virtual screening - efficiently screen "giga-sized" libraries.
• Scaffold hopping - discover structurally novel chemicals based on a lead.
• Hit follow-up - quickly discover new chemicals with similar 3D pharmacophoric binding properties to your lead.

A simple web-based interface is provided for the ease of use by chemists, or can be used in the ICM-Pro software. In either format, RIDE allows fine-tuning of the queries to generate most desirable hit lists. This includes:

• Atom Weighting - Contributions of different portions of the molecule can be modulated with per-atom weights to reflect relative importance of certain moieties.
• Excluded Volumes meet Shape Matching - An envelope penalty can be applied to the regions that surround all or part of the query molecule to prioritize hits without bulky extensions in constrained regions.

13.4.1 RIDE Setup

[ RIDE Server Setup | RIDE GPU Benchmark ]

Prepare the template to search against:

• Convert one or more chemicals to 3D.
• If you are using more than one chemical in your template it the chemicals need to be superimposed.

Download the pre-converted 3D database for screening:

• Pre-converted databases from chemical vendors can be downloaed here: http://molsoft.com/addons/conf/

Setup RIDE:

• Chemistry/APF Tools/ Ride search...
• Use the browse buttom to select the database you have downloaded that you wish to search against the template. In version 3.9-3 you can select multiple databases to screen in one go - use CTRL/Command key to select multiple databases.
• Select the 3D template - e.g. double click on it in the ICM Workspace.
• Enter an APFcutoff. The APFCutoff [0-1] is the percentage of self score of the template. Lower the value - more hits are accepted.
• You can choose to penalize the score if atoms are located outside the envelope. Select a repulsion weight and exclude margin. It penalize within + from template atoms. Note that if you have either envelope penalty or excluded volume you may need to lower the value.
• You can choose to use an excluded volume to represent a pocket. In this case you would need to read in a PDB file and make sure your template molecules are superimposed in the pocket. Then check the excluded volume box and select the pdb structure using the drop down button. If you find chemicals that go into the restricted space this is normal because possibly no chemicals were found with better score but you can increase Repulsion Weight to get more penalty for violating excluded volume.
• "Refine maps": may give you better supposition because the final hits are realigned using maps with smaller step.

How to set atom weights:

You can set weights by changing the Occupancy of each atom from 0-1. By default they are set to 1. To change to 0 or some other value - select the atoms and right click and choose edit Occupancy.

13.4.1.1 RIDE Server Setup

• Chemistry/APF Tools/ Ride search...
• Click on the Server tab

GPU-enabled RIDE server can be installed from Docker container (instructions are contained in download)

http://molsoft.com/distrib/docker/RIDE/

After the Docker is installed -paste the URL into URL box, press refresh button next to the database combo to refresh DB list.

Other parameters are the same as for local search (see above).

13.4.1.2 RIDE GPU Benchmark

MolSoft has tested RIDE on two GPUs:

• Tesla P100 (12Gb RAM) (3 years old model) ~$4K
• GeForce RTX 2080 (8Gb RAM) (last year model) ~$700

Surprisingly, the RIDE performance on RTX was even better ~10%-15% than on the more expensive Tesla. It can be explained because we do not use double precision arithmetic in RIDE superposition, so P100 doesn't give you any advantage and the RTX newer model has higher GPU clock rate.

The algorithm performance linearly depends on the size of your template (number of bonds). 0.5 million conformers/sec performance refers the ligand with ~22 bonds. The GPU RAM required for that ligand size is ~200Mb. For multiple templates everything (speed/RAM requirements) is also scaled linearly.

The other important requirement for the fast RIDE performance is that conformer DB should be located on SSD drive, otherwise the bottleneck will be just reading data from the disk.

In Summary

• For good RIDE performance you don't need expensive professional GPU (like Tesla). Good recent GeForce card will perform even better.
• So, any modern gamer PC will actually work.
• For example we've bought this box some time ago (under $2000) and it works pretty well on RIDE https://www.lenovo.com/us/en/desktops-and-all-in-ones/legion-desktops/legion-t-series-towers/Lenovo-Legion-T730-28ICO/p/99LE9700308?notPassDisplayRuleRedirectFlag=notPassDisplayRuleRedirectFlag