Which Approach is recommended for Fine-Tuning Molecular Property Prediction

[ASIWU]

I am working on fine-tuning uni-mol for predicting molecular properties and have noticed an inconsistency in the data processing methods provided in the repository. There are some small differences between the smi2_3Dcoords function in the example notebook and the inner_smi2coords function in the conformer.py file within the MolTrain class for unimol_tools.

So, which approach is more recommended for fine-tuning unimol for molecular property prediction — example jupyter notebook or the MolTrain in unimol_tools ?

Best regards,

[Naplessss]

smi2_3dcoords is for conformation diversity for docking initial, we prefer to use inner_smi2coords as just use MMFF force sampling coordinates for fientuning.