CFTR_TL: Transfer Learning-Enhanced Prediction of CFTR ATP Binding Sites with Multi-Window Convolutional Neural...

Area of Science:

• Biochemistry
• Molecular Biology
• Computational Biology

Background:

• The Cystic Fibrosis Transmembrane Conductance Regulator (CFTR) protein regulates chloride ion transport and requires ATP binding for function.
• Mutations in CFTR's nucleotide-binding domains (NBDs) are linked to cystic fibrosis, highlighting the need for accurate ATP binding site prediction.
• CFTR, an ATP-binding cassette (ABC) transporter, presents unique challenges for general prediction methods due to its ion channel function.

Purpose of the Study:

• To develop an accurate method for predicting ATP binding sites in CFTR.
• To address the limitations of existing prediction tools for CFTR's unique structural and functional characteristics.
• To improve understanding of CFTR function and facilitate the development of targeted therapies for cystic fibrosis.

Main Methods:

• Developed CFTR_TL, a novel approach using transfer learning for ATP binding site prediction.
• Trained a base model on diverse ATP-binding proteins, then fine-tuned it using data from ATP-binding cassette (ABC) transporters.
• Employed a multiwindow convolutional neural network (CNN) to identify spatial patterns crucial for binding site prediction.

Main Results:

• CFTR_TL demonstrated superior performance compared to traditional prediction methods.
• The model achieved enhanced accuracy and specificity in identifying critical ATP binding residues within CFTR.
• The transfer learning approach effectively leveraged functional similarities within the ABC transporter family.

Conclusions:

• CFTR_TL provides a powerful and accurate tool for CFTR research and drug discovery.
• The method offers a generalizable framework for improving ATP binding site prediction in other protein families.
• Accurate prediction of CFTR ATP binding sites is crucial for advancing cystic fibrosis therapeutics.