search

deepmodeling / Uni-Mol

Official Repository for the Uni-Mol Series Methods
MIT License
726 stars 127 forks source link

构象优化评测问题 #149

Open zhangxiaaobo opened 1 year ago

zhangxiaaobo commented 1 year ago

在构象优化drugs集评测时，使用conf_gen_cal_metrics脚本对评测分子的初始构象生成时，一直跑不出结果，看了具体过程，是先生成M多的初始构象，然后聚类成N类（N为评测构象的2倍），再在每类取一个，共取2N个初始构象，整个初始构象生成过程很耗时，为何要这样操作，这样操作评估效果最好吗？真正在推理过程中，这样操作，会增加推理的时间吧？并且用的target的mol去生成初始构象，为何不是用smiles，这块有点不理解，多谢解答

mode="gen_data" nthreads=20 # Num of threads reference_file="./conformation_generation/drugs/test_data_200.pkl" # Your reference file dir output_dir="./conformation_generation/drugs" # Generated initial data dir

python ./unimol/utils/conf_gen_cal_metrics.py --mode $mode --nthreads $nthreads --reference-file $reference_file --output-dir $output_dir

ZhouGengmo commented 1 year ago

On Drugs, it's indeed slow; you can speed it up by utilizing a multi-CPU machine with increased threads. The quality of initial conformations will impact the results of subsequent evaluations. Despite this, the inference time remains shorter compared to the diffusion model. For the target mol, this approach has been used in previous methods; we align with them."