Generalized Property-Based Encoders and Digital Signal Processing Facilitate Predictive Tasks in Protein Engineering

Area of Science:

• Protein Engineering
• Computational Biology
• Bioinformatics

Background:

• Protein sequence encoding is crucial for computational methods in protein engineering.
• Selecting optimal physicochemical properties for encoding remains a challenge.

Purpose of the Study:

• To generalize property-based encoding strategies for maximizing predictive model performance.
• To develop new encoders that improve accuracy and reduce overfitting in protein engineering tasks.

Main Methods:

• Partitioned the AAIndex database into eight property groups using text mining and unsupervised learning.
• Applied non-linear Principal Component Analysis (PCA) within each group to create representative encoders.
• Assessed model performance using proposed encoders, Fast Fourier Transform (FFT), and classical methods (One Hot Encoder, TAPE embeddings).

Main Results:

• Models trained with proposed encoders and FFT showed significantly increased precision and reduced overfitting.
• The novel encoding strategies outperformed classical methods in predicting protein and peptide function, folding, and biological activity.
• A preliminary methodology for de novo sequence design was proposed.

Conclusions:

• Generalized property-based encoding, combined with FFT, offers a simple yet effective way to enhance predictive tasks in protein engineering.
• These methods improve model performance without increasing computational complexity.
• The approach provides a foundation for designing protein sequences with specific desired properties.