Molecular Descriptors as Continuous Inputs

[CValse]

Hi,
I'm exploring BOFire with continuous molecular descriptors as inputs. I'd like to discuss strategies for handling this scenario, especially when dealing with suggested candidates that may not exist in the set of available molecules.

Is nearest neighbor a viable strategy for evaluating new candidates when the suggested points are not in the set of available molecules?
Given that molecular descriptors are continuous, how can we leverage this continuity while being constrained to a finite set of available structures?
I'm wondering if it's possible to achieve a functionality akin to CategoricalMolecularInput using continuous descriptors.

I'm eager to hear your insights and experiences on this matter. Any relevant papers, code snippets, or best practices would be greatly appreciated! Thanks