PANDA: Predicting the change in proteins binding affinity upon mutations by finding a signal in primary structures

Area of Science:

• Computational biology
• Biophysics
• Machine learning in bioinformatics

Background:

• Accurate prediction of protein binding affinity changes upon mutation is vital for developing novel therapeutics and understanding mutagenesis.
• Current computational methods often rely on protein structures, limiting their applicability to complexes with known 3D structures.
• Assessing the generalization ability of sequence-based predictors requires careful validation, questioning the effectiveness of standard cross-validation (CV) across mutations.

Purpose of the Study:

• To develop and evaluate a sequence-based computational method for predicting changes in protein binding affinity upon mutation.
• To investigate the effectiveness of cross-validation strategies for sequence-based mutation effect predictors.
• To provide a widely applicable and accessible tool for predicting mutation-induced binding affinity changes.

Main Methods:

• Utilized protein sequence information, eschewing protein structures.
• Employed machine learning techniques for prediction.
• Developed a novel sequence-based predictor named PANDA.
• Validated performance using an appropriate CV scheme and an external independent test dataset.

Main Results:

• PANDA demonstrates comparable performance to existing methods when assessed with a rigorous CV scheme and an independent test set.
• On an external test dataset, PANDA achieved a maximum Pearson correlation coefficient of 0.52.
• The state-of-the-art structure-based method, MutaBind, achieved a maximum Pearson correlation coefficient of 0.59 on the same dataset.

Conclusions:

• Sequence-based prediction of binding affinity changes upon mutation is feasible and offers wide applicability.
• PANDA provides a valuable alternative to structure-based methods, especially when 3D structures are unavailable.
• The developed method, PANDA, is accessible via a cloud-based webserver and Python code, facilitating its use in research.