Dryad helps drug discovery scientists leverage the power of AI/ML without requiring a team of data scientists. Through the power of Wisecube’s AI platform Nephos, scientists can not only utilize the 30 out of the box ADMET models but also train their own custom models with a click of a button.
This article will walk you through the process of training your own custom Drug discovery models.
The first step is to click on the ‘Train’ tab of the Dryad application which should take you to a listing screen like so:
Now go ahead and click the ‘Train Model’ button to get started. You should be presented with a modal that says Add a new model’ like so:
Upload a set of compounds similar to what you did in the ‘Analyze’ section with the additional column included that is your target variable. The target variable is the value that you want the AI model to learn how to predict. In effect there should at least be two columns in the file uploaded like so:
In the above example the new column ‘WT Inhibition (%)’ is the target variable the Dryad application is asked to learn how to predict, using the SMILES column as the only input. Once you upload the CSV file containing data similar to the above example you should be taken to the next stage in the model training workflow like so:
Once you select the SMILES column and the Target field column you can go ahead and click on the ‘Create Model’ button to move the workflow to the next stage. In this stage you will be asked to name your new custom AI model and give it a description for future reference like so:
Once you have done that you can move to the next stage in the model training process and select the training workflow to use to train your custom model. If you know which specific training workflow to use here, go ahead and select it, if not reach out to your Account manager at [email protected] to confirm which one to use.
The final step of the model training process is to confirm the preview screen to ensure you have all the right selections before you kickoff your model training workflow like so:
If everything is looking good, go ahead and click on the ‘Start Train model’ button to start the model training workflow, if not feel free to go back and click on ‘Change workflow’ button to change aspects of the model training.
Once you click the ‘Start Train Model’ button you should see the listings page again with a new row that says ‘preparing’ and should indicate the progress of the workflow itself like so:
Eventually the mode should finish training and change to ‘Succeed’ status and be ready to be used in a new analysis run.