Partial diffusion

[adrienchaton]

Dear all,

This issue has some relation to #1

I understand that you are working on different ways to apply and condition DrugFlow, as a first step, I imagine that it should be able to support partial diffusion in its present form, and I would ask if you may be able to provide some example script / guidance on how to implement that please.

The use-case is instead of running 5k diffusion steps from a randomly initialised atom distribution, take a reference ligand, noise for X steps and denoise for X steps. This should already allow experimenting with molecular optimisation / diversification and coarse conditioning... i.e. low X should give strong input constrain and higher X should tend towards pure de-novo.

Is this something that you have experimented with? Or am I wrong and that cannot be done with the current model version?

Thanks for sharing your great work!