CysPresso: a classification model utilizing deep learning protein representations to predict recombinant expression...

Area of Science:

• Biotechnology
• Computational Biology
• Drug Discovery

Background:

• Cysteine-dense peptides (CDPs) are promising pharmaceutical scaffolds due to their biochemical properties, low immunogenicity, and high target affinity.
• Chemical synthesis of CDPs is challenging, making recombinant expression a viable alternative.
• Predicting CDP expression in mammalian cells is crucial for gene and mRNA therapy applications but currently lacks efficient methods.

Purpose of the Study:

• To develop a computational model for predicting the recombinant expression of CDPs in mammalian cells.
• To identify suitable protein representations for accurate expression prediction.
• To facilitate the identification of expressible CDPs for therapeutic development.

Main Methods:

• Developed CysPresso, a machine learning model utilizing primary sequence data.
• Evaluated deep learning-derived protein representations (SeqVec, proteInfer, AlphaFold2).
• Optimized the model using AlphaFold2 representations, time series transformation with random convolutional kernels, and dataset partitioning.

Main Results:

• AlphaFold2 representations demonstrated the best predictive performance for CDP expression.
• The optimized CysPresso model accurately predicts recombinant CDP expression in mammalian cells.
• Random convolutional kernel transformation was superior to embedding averaging for preserving relevant information.

Conclusions:

• CysPresso is the first model to predict recombinant CDP expression in mammalian cells, particularly for knottin peptides.
• Deep learning-based protein representations, like AlphaFold2, are applicable beyond structure prediction.
• The study highlights the potential of computational tools in accelerating therapeutic peptide development.