Deep Learning for Predicting Protein Interactions
Author Information
Author(s): Majila Kartik, Ullanat Varun, Viswanath Shruthi
Primary Institution: National Center for Biological Sciences, Tata Institute of Fundamental Research, Bangalore, India
Hypothesis
Can a deep learning method accurately predict interactions for intrinsically disordered regions of proteins?
Conclusion
The Disobind method outperforms existing state-of-the-art methods in predicting inter-protein contact maps and interface residues for intrinsically disordered regions.
Supporting Evidence
- Disobind outperforms AlphaFold-multimer and AlphaFold3 at multiple confidence cutoffs.
- Combining Disobind and AlphaFold-multimer predictions further improves performance.
- Disobind achieves an F1-score of 0.57 on the ID test set for contact map predictions.
Takeaway
This study created a new tool called Disobind that helps scientists understand how certain proteins interact, especially those that don't have a fixed shape.
Methodology
Disobind uses a deep learning model trained on a dataset of merged binary complexes to predict inter-protein contact maps and interface residues from protein sequences.
Potential Biases
The predictions may be biased due to the limited number of experimental structures available for IDRs.
Limitations
The method is limited to binary IDR-partner complexes and requires input sequences to be less than 100 residues long.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website