ProtSATT: An Advanced Protein Solubility Predictor Based on Attention Mechanism

Area of Science:

• Biotechnology
• Computational Biology
• Protein Engineering

Background:

• Protein solubility is critical for biologics production but often limited by aggregation.
• Current optimization methods are empirical and time-consuming.
• Existing computational tools do not fully leverage modern protein language models (PLMs).

Purpose of the Study:

• To develop ProtSATT, a sequence-based computational framework for predicting protein solubility.
• To integrate complementary information from multiple PLMs for improved prediction accuracy.
• To provide a computationally efficient tool for protein solubility prediction in biologics development.

Main Methods:

• ProtSATT integrates embeddings from UniRep, ESM-2, and ProtT5 PLMs.
• Attention-based feature extraction and fusion are applied in the latent space of embeddings.
• The framework is evaluated on solubility regression and expression-related classification tasks across three benchmarks.

Main Results:

• ProtSATT achieved competitive performance on eSOL (R^2=0.5450, accuracy=81.21%) and S. cerevisiae (accuracy=83.33%).
• On the E. coli benchmark, it showed competitive expression classification accuracy (72.21%).
• The model is computationally efficient, processing >11,400 sequences/sec with 6.0M parameters.

Conclusions:

• Integrating multiple PLM representations with attention-based modeling enhances protein solubility prediction.
• ProtSATT offers a computationally efficient solution for protein engineering and biologics development.
• The framework provides valuable computational support for optimizing protein production and utility.