Post-translational modification prediction via prompt-based fine-tuning of a GPT-2 model

Area of Science:

• Biochemistry
• Computational Biology
• Genomics

Background:

• Post-translational modifications (PTMs) are crucial for protein function, impacting cellular signaling, localization, and degradation.
• Predicting PTMs is complex due to intricate biological interactions, necessitating advanced computational methods.

Purpose of the Study:

• To introduce PTMGPT2, an interpretable protein language model designed for accurate PTM site prediction.
• To leverage prompt-based fine-tuning and unsupervised learning for enhanced PTM identification.

Main Methods:

• PTMGPT2 utilizes a custom prompt to guide unsupervised learning on amino acid sequences, identifying PTM sites.
• Interpretability is achieved through visualizing attention profiles and analyzing mutation effects on PTM sites.

Main Results:

• PTMGPT2 demonstrates superior performance compared to existing methods across 19 distinct PTM types.
• The model effectively identifies sequence motifs critical for molecular recognition and protein functionality.

Conclusions:

• PTMGPT2 offers a powerful and interpretable approach for PTM prediction.
• The model holds significant potential for identifying disease associations and novel drug targets.