DG-PBs: A web server for diffusion-enhanced protein–DNA binding site prediction under class imbalance

Residue-level prediction tool based on diffusion-enhanced GNN and cosine similarity Top-12 graph construction

Submit Job Query Results About

Submit Prediction Job

Supported file types: .fasta, .fa, .faa, and .txt. The file may contain one or multiple FASTA records.
Only protein sequences are required. True 0/1 labels are not needed. If a pure 0/1 label line is pasted by mistake, it will be ignored automatically.
If the server email service is configured, a completion notification will be sent to this address.
This helps identify your job. A unique job ID will still be generated automatically.
Residues with predicted probability >= threshold are labeled as DNA-binding residues. The default threshold is 0.50.
Waiting for submission. The first prediction may take longer because ESM-2 and the GNN model need to be loaded.

Job Status

After submission, the job ID, current status, prediction statistics, and download buttons will be shown here.

Query Results

Enter a job ID to query saved prediction results.
As long as the job directory is still stored on the server, results can be queried and downloaded by job ID.

Query Result

The queried job information will be shown here.

Recent Jobs

Job ID Status Sequences Created Time Action
job_20260610_160007_de4f8db2 Finished 3 2026-06-10 16:00:07 View
job_20260610_152559_dcca4220 Finished 2 2026-06-10 15:25:59 View
job_20260610_143637_b498cada Finished 2 2026-06-10 14:36:37 View
job_20260610_142957_824b53c3 Finished 2 2026-06-10 14:29:57 View
job_20260610_142947_16a9fb64 Finished 2 2026-06-10 14:29:47 View
job_20260610_111021_1852fc5c Failed 2 2026-06-10 11:10:21 View
job_20260610_110649_78363b7f Failed 2 2026-06-10 11:06:49 View
job_20260610_094210_6bbdab15 Finished 2 2026-06-10 09:42:10 View

About This System

This system is an online protein-DNA binding site prediction platform. It converts the trained diffusion-enhanced graph neural network model into a web-based tool. Users only need to upload or paste protein FASTA sequences, and the system will return residue-level DNA-binding probabilities and downloadable prediction results.

Input

Users can submit one or multiple protein sequences in FASTA format. True labels are not required for online prediction.

Prediction Pipeline

The server extracts residue-level ESM-2 embeddings, constructs a cosine similarity Top-12 residue graph, and performs GNN inference.

Output

The system provides a job ID, prediction status, residue-level CSV results, sequence summary, and optional email notification.

System Workflow

FASTA sequence submission → job ID generation → background prediction → ESM-2 feature extraction → cosine similarity graph construction → GNN inference → result storage → query and download.

Cites

Diffusion and Graph Neural Network-based Protein–DNA Binding Site Prediction
Hanqing Zhang, Weisen Yang, Weizhong Lin+
School of Information Engineering, Jingdezhen Ceramic University, Jingdezhen 333403, Jiangxi, China

Contact Person: Weizhong Lin
Correspondence Address (Postal Code): Jingdezhen Ceramic University 333403
E-mail: linweizhong@jcu.edu.cn
沪ICP备2026044284号-1