Benchmark for SeqScreen

The paper is under review.

[Github Repo] | [Model Collection] | [Cite]

Abstract

Virtual screening aims to identify candidate molecules that bind to a target protein, playing a central role in computational drug discovery. Sequence-based deep learning methods offer an applicable alternative to structure-based approaches, but typically process one protein-molecule pair at a time, limiting their scalability to large molecular libraries. Contrastive learning methods inspired by CLIP have shown promise for learning joint protein-molecule representations, but standard CLIP training was designed for symmetric tasks and does not account for the asymmetric and one-to-many nature of protein-molecule binding. In this paper, we introduce SeqScreen, a sequence-based virtual screening method built on a dual-encoder contrastive architecture. SeqScreen introduces a protein-centric batch construction strategy and an asymmetric multi-positive InfoNCE loss to cope with the protein-centric nature of virtual screening. We conduct a systematic evaluation across 8 protein language models and 3 molecular language model variants. The protein-centric batch construction consistently outperforms standard CLIP training across all evaluated encoders, while requiring approximately 32 times fewer training epochs and 7 times fewer forward passes during inference compared to pair-based methods. On the LIT-PCBA dataset, SeqScreen outperforms all sequence-based baselines, achieving a relative improvement of up to 39% in EF at 0.5 over the best competing method, while remaining competitive with traditional docking approaches without requiring 3D structural information.

Dataset Details

ChEMBL

The dataset consists of 1,492 proteins, 394,190 molecules, and 583,960 positive pairs for training; 334 proteins, 101,586 molecules, and 147,676 positive pairs for validation; 326 proteins, 105,048 molecules, and 130,281 positive pairs for testing. The dataset was generated using ChEMBL 36, split by protein sequence similarity.

LIT-PCBA

The dataset consists of 1,452 proteins, 375,120 molecules, and 548,294 positive pairs for training; 326 proteins, 97,213 molecules, and 143,088 positive pairs for validation; 15 proteins, 404,586 molecules, and 2,776,973 pairs (positives and negatives) for testing. The training and validation datasets were generated using a filtered version of the ChEMBL dataset. The test set is the LIT-PCBA dataset.

Dataset Usage

ChEMBL

Use the code below to load the ChEMBL dataset.

from datasets import load_dataset 

dataset_dict = load_dataset("SaeedLab/SeqScreen", data_dir="chembl")

LIT-PCBA

Use the code below to load the LIT-PCBA dataset.

from datasets import load_dataset 

dataset_dict = load_dataset("SaeedLab/SeqScreen", data_dir="lit_pcba")

Citation

The paper is under review. As soon as it is accepted, we will update this section.

License

This model and associated code are released under the CC-BY-NC-ND 4.0 license and may only be used for non-commercial, academic research purposes with proper attribution. Any commercial use, sale, or other monetization of this model and its derivatives, which include models trained on outputs from the model or datasets created from the model, is prohibited and requires prior approval. Downloading the model requires prior registration on Hugging Face and agreeing to the terms of use. By downloading this model, you agree not to distribute, publish or reproduce a copy of the model. If another user within your organization wishes to use the model, they must register as an individual user and agree to comply with the terms of use. Users may not attempt to re-identify the deidentified data used to develop the underlying model. If you are a commercial entity, please contact the corresponding author.

Contact

For any additional questions or comments, contact Fahad Saeed (fsaeed@fiu.edu).