RTK_RAG: Leveraging Retrieval Augmented Generation with Multi-Window Convolutional Neural Networks for Superior ATP...

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Cancer Research

Background:

• Receptor tyrosine kinases (RTKs) are crucial for cell signaling and implicated in cancer.
• Accurate identification of ATP binding sites is essential for understanding RTK function and developing targeted therapies.
• Existing general ATP binding site predictors struggle with the structural diversity of RTKs.

Purpose of the Study:

• To develop an advanced computational framework, RTK_RAG, for improved prediction of ATP binding sites specifically in RTKs.
• To address the limitations of general prediction methods in capturing RTK-specific structural variations.
• To provide a reliable tool for RTK research and kinase inhibitor development.

Main Methods:

• Integration of retrieval-augmented generation (RAG) with protein language models (PLMs).
• Utilization of a multi-window convolutional neural network (MCNN) architecture.
• Validation on an independent RTK dataset using multiple evaluation metrics.

Main Results:

• RTK_RAG demonstrated superior performance compared to general ATP binding site predictors on RTKs.
• The framework effectively accounts for RTK-specific structural features, enhancing prediction accuracy.
• Achieved improved performance across multiple evaluation metrics on an independent RTK dataset.

Conclusions:

• RTK_RAG offers a reliable tool for studying RTK function and advancing the development of novel kinase inhibitors.
• The RAG-based approach shows promise for improving functional predictions in specialized protein families.
• This study presents a generalizable strategy for enhancing binding site identification across diverse protein families.