ESMDisPred: A Structure-Aware CNN-Transformer Architecture for Intrinsically Disordered Protein Prediction.

Area of Science:

• Computational Biology
• Structural Biology
• Biochemistry

Background:

• Intrinsically disordered proteins (IDPs) lack stable structures but are crucial for biological processes.
• IDPs' flexibility complicates experimental studies and links them to diseases like cancer.
• Accurate computational prediction of IDPs is vital for research, drug discovery, and protein engineering.

Purpose of the Study:

• Introduce ESMDisPred, a novel structure-aware predictor for intrinsically disordered proteins.
• Enhance the accuracy of computational disorder prediction by integrating sequence and structural information.
• Improve understanding of IDP roles in biological systems and disease.

Main Methods:

• Utilized Evolutionary Scale Modeling-2 (ESM2) protein language models for sequence embeddings.
• Integrated sequence embeddings with structural data from the Protein Data Bank (PDB).
• Employed a hybrid CNN-Transformer architecture with feature engineering strategies.

Main Results:

• ESMDisPred achieved state-of-the-art prediction accuracy on CAID3 benchmarks.
• Achieved ROC-AUC of 0.895, AP of 0.778, and a max F1 score of 0.759.
• Demonstrated superior performance compared to recent disorder prediction methods.

Conclusions:

• Integrating protein language model embeddings with explicit structural information improves disorder prediction.
• ESMDisPred offers a powerful tool for studying IDPs.
• This approach advances computational biology and structural biology research.