Can you provide a toml example on how to use Qptuna trained models to further generate new molecules?

[MachineGUN001]

Can you provide a toml example (*.toml) on how to use example Qptuna trained models to further generate new molecules?

how can I set the config for QSAR/QSPR model-related components, eg Qptuna: QSAR models with Qptuna?
as for mode_file: model file name ,
How do I set up this model file in the config file? The model file here would be the trained model via Qptuna, right?

many thanks

[halx]

A Qptuna configuration is just like any other, see below.

[[component.Qptuna.endpoint]]
name = "Qptuna"
weight = 0.6

params.model_file = "merged.pkl"  # or "best.pkl" as needed

# adjust transform as needed
transform.type = "reverse_sigmoid"
transform.high = 0.0
transform.low = -50.0
transform.k = 0.4

[MachineGUN001]

thanks for your help.

Is it necessary to set up in the tmol configuration file what type of model is being trained: regression model or classification model?
I set it again in scoring.tmol according to the configuration conditions you provided, and the scoring gives a result of 1 or 0, which looks as if REVINVENT recognizes it as a classification model.

Many many thanks,

[halx]

You will need to give more details as to what "did not work" means. You will have to adjust the transform to the dynamic range of your model output and how you want to have that transformed.

[halx]

REINVENT does not recognize what type of model you, it is rather you who has to sort this out. The mechanim for that is the tranformation function. If you have a binary classification model that also outputs only 0s and 1s you do not need a transform unless you need to invert the model's response.

The TOML snippet above assumes that you use the include mechanism in your main TOML file.

[stage.scoring]
type = "geometric_mean"
filename = "scoring.toml"
filetype = "TOML"

If you use this in your main TOML file you would have to prefix all section with stage.scoring. Whether that is for stage1 or stage2 is for you to decide.

[MachineGUN001]

got it. thanks a lot!