Protein Large Language Models Can Predict Flavivirus Protease Target Specificity.

Area of Science:

• Virology
• Computational Biology
• Biochemistry

Background:

• Viral proteases are critical for viral replication and pathogenesis.
• Flavivirus proteases display substrate promiscuity, complicating drug development.
• Understanding protease-substrate interactions is key for antiviral strategies.

Purpose of the Study:

• To investigate the substrate specificity of five flavivirus proteases: Yellow Fever virus protease (YFP), West Nile Virus Protease (WNP), Zika virus protease (ZVP), Usutu Virus Protease (UVP), and Rocio Virus Protease (RVP).
• To explore the utility of protein language models (pLMs) in predicting viral protease-peptide interactions.

Main Methods:

• Recombinant expression and purification of five flavivirus proteases.
• Mass spectrometric Proteomic Identification of protease Cleavage Sites (PICSs) using peptide libraries.
• Homology modeling to identify protease subsites.
• Fine-tuning the ProtTrans T5-Encoder protein language model with experimental peptide data.

Main Results:

• All five flavivirus proteases demonstrated significant substrate promiscuity, with a preference for arginine at the P1 position.
• Homology modeling provided insights into subsites but struggled to fully explain binding promiscuity.
• The fine-tuned ProtTrans T5-Encoder model achieved up to 76% accuracy in classifying peptide interactions.
• T5 embeddings captured relevant information for recognizing protein-peptide interactions.

Conclusions:

• Flavivirus proteases exhibit unexpected substrate promiscuity.
• Protein language models, specifically ProtTrans T5-Encoder, are effective tools for predicting viral protease-peptide interactions.
• These findings have significant implications for the rational design of novel antiviral drugs targeting viral proteases.