ΔΔG Prediction Engine
DeepPNI

Mutation-driven protein–nucleic acid binding free energy prediction using protein language model and relational graph convolution network

01Upload CSV
02Build Graphs
03ESM-2 Embed
04DeepPNI Predict
05Results
↑Input Mutations
🧬
Drop your mutation CSV here
or click to browse  ·  .csv files only  ·  max 5 KB
📄
Please fix:
    Required CSV Format
    mutationPDB_IDchain
    Y596A6so9A
    V537I6so9A
    R595M6so9A

    Mutation format: WT + position + MUT e.g. Y596A  · 

    ⟳Pipeline Status
    Running
    Initialising…
    ≡Predictions
    Mutations Predicted
    —
    Mean ΔΔG (kcal/mol)
    —
    Range (kcal/mol)
    —
    PDB IDChainMutation Predicted ΔΔG (kcal/mol)Binding stability
    Download CSV
    ◎3D Structure Viewer — Mutation Map

    Click any result row to load and highlight that mutation on the 3D structure. Use the toolbar to change style, colour scheme, and appearance.

    Style
    Colour
    Show
    ⚙
    ↺
    ⬜
    🔬
    Click a result row to load its structure
    RCSB PDB · NGL Viewer · mutations highlighted
    Loading…
    Opacity
    Protein92%
    Nucleic95%
    Mutations90%
    Protein colour
    Nucleic colour
    No structure loaded
    Protein
    Nucleic acid
    Destabilising
    Stabilising

    Drag to rotate  ·  Scroll to zoom  ·  Right-click drag to translate  ·  Double-click to centre  ·  ⚙ for appearance options

    Scientific Background

    Protein–Nucleic Acid Interactions

    Protein–nucleic acid interactions have influence in numerous biological functions, such as gene expression, DNA replication, repair, translation, and recombination. The effect of amino acid mutation in the binding affinity between protein and nucleic acid is focused in this study. Here we built a regression model to study effect of amino acid mutation on the binding free energy change (ΔΔG). A positive ΔΔG indicates destabilising mutation and a negative ΔΔG indicates stabilising mutation.

    DeepPNI Architecture

    Our DeepPNI model combines sequence and structural information of protein-nucleic acid complex for the ΔΔG prediction. We have used (a) ESM-2 protein language model to capture the sequence information and (b) RGCN to extract 3D structural information at the local environment of the mutating residue. The combined feature is passed through a three-layer MLP for the ΔΔG prediction.

    Dataset Information

    We trained our DeepPNI model on 1754 mutational data from NABE database. For external validation we have used mutational data from ProNAB database. For more details about the datasets, go to our main paper.

    Tutorial

    A step-by-step example is provided for users using publicly available PDB structures. Results can be obtained within minutes.

    Example — Multi-PDB mutation panel (PDBs: 2C4R, 3IMB, 3KMD, 1AZ0)
    1

    Download the demo CSV

    We provide a ready-to-use demo file containing 7 mutations across 4 different PDB structures. Click the button below to download it directly:

    ↓ Download demo.csv

    The file contains the following mutations:

    demo.csv
    mutationPDB_IDchain
    D346N2c4rL
    D303N2c4rL
    H219N3imbD
    H77A3imbD
    C124S3kmdD
    C141S3kmdD
    K38A1az0A
    2

    Upload and run

    Navigate to the Predictor tab. Drag and drop demo.csv onto the upload zone, or click to browse and select it. Then click Run Prediction. The progress bar advances through five pipeline stages, typically taking 1–2.5 minutes for 100–350 mutations.

    3

    Interpret results

    The results table shows the predicted ΔΔG (kcal/mol) for each mutation. Positive ΔΔG indicates a destabilising mutation (weakens binding); negative ΔΔG indicates a stabilising mutation (strengthens binding).

    4

    View mutations on the 3D structure

    Click any result row — the structure loads automatically in the 3D viewer. Destabilising mutations appear as red spheres; stabilising as blue spheres. Users can change the representation in the visualiser. Use drag/scroll to explore the structure. Users can download the snapshot of the representation for further use.

    5

    Download your results

    After prediction, click ↓ Download CSV. The output file contains all input columns plus a predicted_ddg column. Import into Python (pandas) or Excel for downstream analysis.

    Help

    Input File Format

    Upload a comma-separated (.csv) file with exactly three columns: mutation, PDB_ID, chain. Column headers are case-sensitive. Mutation format: WTpositionMUT (e.g. Y596A). Only single-point mutations are supported.

    Interpreting ΔΔG

    ΔΔG = ΔG(mutant) − ΔG(wild-type). Positive ΔΔG = mutation weakens binding (destabilising). Negative ΔΔG = mutation strengthens binding (stabilising). Units are kcal/mol.

    3D Structure Viewer

    The viewer uses NGL Viewer — the same rendering engine used by UniProt, PDBe, and RCSB PDB. Click any result row to automatically load and highlight the mutation site. Protein chains are coloured in sage green (#7a9e7e) and nucleic acids in navy blue (#1c3a6e).

    Citation

    If you use DeepPNI in your research, please cite the work below.

    Primary Reference

    Mondal,S., Mondal,T., Pramanik,S. and Mehra,R. (2025) DeepPNI: a language-and graph-based model for mutation-driven protein-nucleic acid binding energetics. Nucleic Acids Res 2026;:gkag820. https://doi.org/10.1093/nar/gkag820

    BibTeX

    @article{Mondal2025DeepPNI, author = {Mondal, S. and Mondal, T. and Pramanik, S. and Mehra, R.}, title = { DeepPNI: a language-and graph-based model for mutation-driven protein-nucleic acid binding energetics}, journal = {Nucleic Acids Research}, year = {2026}, doi = {https://doi.org/10.1093/nar/gkag820}, note = {} }

    Preprint

    Available on arXiv: arXiv:2511.22239 ↗

    Contact Us

    Send a Message
    Contributors
    SM
    Somnath Mondal
    somnath@iitbhilai.ac.in
    TM
    Tinkal Mondal
    tinkalm@iitbhilai.ac.in
    SP
    Dr. Soumajit Pramanik
    soumajit@iitbhilai.ac.in
    RM
    Dr. Rukmankesh Mehra
    rukmankesh@iitbhilai.ac.in
    Team
    Principal Investigator
    IIT Bhilai, Dept. of Chemistry
    ✉ rukmankesh@iitbhilai.ac.in
    Dr. Soumajit Pramanik
    Co-Investigator
    IIT Bhilai, Dept. of Computer Science
    ✉ soumajit@iitbhilai.ac.in
    Technical Support
    Server & Bug Reports
    For technical issues, use the message form or email us directly.
    ✉ molinfo@iitbhilai.ac.in
    📍 IIT Bhilai — Location
    Our Address Office: 307 SD-1, Department of Chemistry
    Indian Institute of Technology Bhilai,
    6th Lane Road, Jevra,
    Bhilai — 491001, Chhattisgarh, India
    ✉ molinfo@iitbhilai.ac.in

    Admin Dashboard

    Admin Sign In