DFT encoding of Substance Parameters

[ustojiljkoff]

Hello!

I am having a rough time finding this in the docs/code/issues, so I thought to open a new one.

Is there a way to define a SubstanceParameter (e.g. ligand) and encode it in the BO using DFT-based descriptors that are previously calculated (and ideally passed as a csv/dictionary)? For example, could one define as a Substance Parameter a series of phosphine ligands, and use descriptors from KRAKEN library for their encoding?

Thanks!

[Scienfitz]

Hi @ustojiljkoff

what you are looking for is called CustomDiscreteParameter

import pandas as pd
from baybe.parameters import CustomDiscreteParameter

descriptors = pd.DataFrame(
    {
        "Kraken_Feature_1": [20, -71, -39],
        "Kraken_Feature_2": [120, 32, 241],
    },
    index=["Molecule A", "Molecule B", "Molecule C"],  # put labels in the index
)

CustomDiscreteParameter(
    name="Ligand",
    data=descriptors,
    active_values=(  # optional, enforces that only Molecule A or C is recommended
        "Molecule A",
        "Molecule C",
    ),
    decorrelate=True,  # optional, uses default correlation threshold, bool or float
)

[ustojiljkoff]

Perfect, works like a charm, thanks!