You've set up a generator, clicked on Run, waited, and yet the generator stops with no (or few) molecules found: what comes next?
Check the exit vector
Makya algorithm is chemistry-driven; therefore, any specified exit vector needs to be an appropriate reactive center.
For more information, see the corresponding article in the documentation.
Check the chemical constraints
Makya lets you specify in great detail the characteristics of the molecules you want to generate. In your generation set-up, in the Products tab, you can indicate:
- Ranges of desired values for the physicochemical properties, such as molecular weight, number of rings, LogP...
- Required or forbidden chemical structures (see the article in the documentation);
- PAINS/Tox dictionaries of forbidden substructures.
Illustration from the Products tab: physicochemical descriptors (left) and substructure filters (right).
The more information the user specifies in this tab, the more generation is constrained, and the more likely the risk of a conflict between the specified requirements and the chemical space available for exploration.
If your generator fails at finding molecules, there might be a conflict with the requirements in the Products tab: check all the information that is specified there and try relaxing some of the constraints.
In the example below, for example, the starting fragment for the generation has 2 aromatic rings (see the greyed column); however, the user specified that the generated molecules should have ≤ 1 aromatic ring (see the Max column). Therefore, the generator will not be able to find solutions.
Check the reward setup
Rewards are scores that are computed on generated molecules and sent back to the generator, which learns from these scores what did and didn't work, and tries to generate new molecules with better scores using a method known as Reinforcement Learning.
Rewards greatly impact the availability of the generator to explore a wide-enough chemical space to find good solutions.
For more information, see the corresponding articles in the documentation:
- Generating with multiple rewards;
- Generating with QSAR models;
- Generating with 3D-LB rewards;
-
Generating with 3D-SB rewards.
Restart the generator
Finally, in order to use computational resources efficiently, Makya implements an automatic stop feature on generators that prevents generators from getting endlessly stuck. Generators that have stopped can be restarted to keep exploring, increasing the chances to find solutions to your problems.
If your generator finds no molecule, and if you have checked the points above to make sure the generator setup is correct, you can restart the generation by clicking on the spinning arrows button:
You will be asked which restart option you want to use:
- Resume from where the generator left off: the generator keeps exploring in the same direction. This option is best if your generator already found some results, and you want to find more.
- Delete the previous state and start from scratch: the generator explores in a different direction. This option is best if the generator found no result, and you want to increase your chances.
In both cases, previously found molecules will be kept in the generator results.