search

ersilia-os / 3d-analogues

Evaluation of analogues in the 3D Space
GNU General Public License v3.0
0 stars 0 forks source link

🐅 Epic: Test new generative models #3

Open GemmaTuron opened 3 months ago

GemmaTuron commented 3 months ago

Summary We used ChemSampler for a first round of generation of dCA analogues, which were evaluated by docking scores only. We still want to get more diversity of compounds so we have been working on incorporating more generative models. We have also identified other ways of evaluating the results, mainly via 3D Shape analysis (see this repository. We have discarded the option of training a bioactivity predictor due to the lack of data on HIV-TAT inhibition. Now we want to evaluate how these new generative models perform and increase the similarity searches to 3D space. By model testing it means:

Objectives

Documentation Previous iterations of chemsampler: https://ersilia.gitbook.io/ersilia-book

GemmaTuron commented 3 months ago

@miquelduranfrigola and @DhanshreeA Can you confirm the status of the following models:

miquelduranfrigola commented 2 months ago

eos8vud and eos69e6 are WIP

GemmaTuron commented 2 months ago

Hi @miquelduranfrigola

This week you confirmed eos69e6 was done. Can you please clarify?

miquelduranfrigola commented 2 months ago

The model is done - I just said "WIP" since I assumed it did not work for you, so I wanted to check what might be going on. But as far as I am concerned, it works.

GemmaTuron commented 2 months ago

Ok, I have tested the following models using dCA as a seed molecule: eos6ost, eos57bx, eos69e6, eos6ru3 and eos8bhe (this last one did not produce results). They generate between 100 - 1000 molecules each, so I am now working on the best way to evaluate those molecules, as docking is very slow for that purpose. See #4 for more.