MolPIF

MolPIF is a Parameter Interpolation Flow model for molecule generation. PIF is a novel generative modeling framework proposed in this work. MolPIF supports pocket-specific molecule generation for de novo and lead optimization tasks.

Technical details and evaluation results are provided in our paper:

• MolPIF: A Parameter Interpolation Flow Model for Molecule Generation

Table of Contents

• MolPIF
  • Table of Contents
  • Installation
  • Prepare Dataset
  • Model weights
  • Training
  • Inference
  • Evaluation
  • License
  • Citation

Installation

You can build the environment (Default CUDA version is 12.4) using:

./setup_env.sh

To activate the environment, run:

conda activate MolPIF

Prepare Dataset

We use the same data as TargetDiff. Data used for training / evaluating the model should be put in the ./data folder by default, and accessible in the data Google Drive folder.

To train the model from scratch, download the lmdb file and split file into data folder:

• crossdocked_v1.1_rmsd1.0_pocket10_processed_final.lmdb
• crossdocked_pocket10_pose_split.pt

To evaluate the model on the test set, download and unzip the test_set.zip into ./data folder. It includes the original PDB files that will be used in Vina Docking.

Model weights

Download the pretrained checkpoint and config from Google Drive whose filenames are pretrained.ckpt and config.yaml, and put it into ./weights folder as follows. You can use the pretrained weight for inference.

• 📂 weights
  • 📂 checkpoints
    • 📄 pretrained.ckpt
  • ⚙️ config.yaml

Training

To train MolPIF, firstly make sure you have prepared the dataset according to Prepare Dataset, and put it in the right folder. it is optional to modify ./configs/default.yaml. After this, you can run:

python train.py

And you will get the intermediate results and the checkpoints in ./logs.

Inference

As an example (Make sure checkpoints are put in the right folder), you can run :

python sample_for_pocket.py --ckpt_path weights/checkpoints/pretrained.ckpt

And you can get results in ./example/output_test/frag_part_filter  

To generate molecules for de novo task targeting specified protein pocket, run:

python sample_for_pocket.py --protein_path $protein_path --ligand_path $ligand_path --ckpt_path $ ckpt_path --out_fn $out_fn

And you will get the results in $out_fn.

To generate molecules for lead optimization task targeting specified protein pocket (recommend using pretrained_lead.ckpt, which is trained with a larger $Pm$ and $Pam$), you need to specify an additional parameter fix_index to indicate the indices of the fixed atoms for the ligand, which can be determined using ./test/get_ligand_index.py. Then run:

python sample_for_pocket.py --protein_path $protein_path --ligand_path $ligand_path --ckpt_path $ ckpt_path --out_fn $out_fn --fix_index $fix_index --attachment_atoms $attachment_atoms --min_add_num $min_add_num  

And you will get the results in $out_fn. ($attachment_atoms is used to specify the anchor, and remove atoms added at undesired positions)

Evaluation

For regular properties (vina score, QED, SA, SE, etc), it is calculated upon sampling. The other evaluation procedure is the same as MolCRAFT and CBGBench; please refer to them for details. For tsne evaluation, you can use ./test/morgan_tsne.py. For toy dataset evaluation, you can refer to ./toy/.

Generated results MolPIF_vina_docked.pt, MolPIF_metrics.json and MolPIF_geom.xlsx can be downloaded from Google Drive

License

This project is licensed under the terms of the GPL-3.0 license.

Citation

@article{jin2026molpif,
  author = {Jin, Yaowei and Wang, Junjie and Tang, Yufan and Xiang, Wenkai and Cao, Duanhua and Teng, Dan and Fan, Zhehuan and Xiong, Jiacheng and Sheng, Xia and Zeng, Chuanlong and An, Duo and Zheng, Mingyue and Zheng, Shuangjia and Shi, Qian},
  title = {MolPIF: a parameter interpolation flow model for molecule generation},
  journal = {Bioinformatics},
  volume = {42},
  number = {6},
  pages = {btag323},
  year = {2026},
  month = {06},
  issn = {1367-4811},
  doi = {10.1093/bioinformatics/btag323},
  url = {https://doi.org/10.1093/bioinformatics/btag323}
}