CycleDNN - A Novel Deep Neural Network Model for CETSA Feature Prediction cross Cell Lines.

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Cellular Thermal Shift Assay (CETSA) is crucial in drug discovery and biological research.
• Conducting CETSA across diverse cell lines is resource-intensive and time-consuming.
• Predicting CETSA features across cell lines can significantly streamline research.

Purpose of the Study:

• To develop a computational method for translating CETSA features between different cell lines.
• To enable automatic prediction of CETSA profiles for proteins in new cell lines.
• To reduce the experimental burden of CETSA.

Main Methods:

• Proposed a novel deep neural network model, CycleDNN, inspired by image-to-image translation techniques.
• CycleDNN utilizes two auto-encoders for cyclic feature translation between cell lines (A to B and B to A).
• Employed reconstructed loss, cycle-consistency loss, and latent vector regularization for model training.

Main Results:

• Demonstrated the effectiveness of CycleDNN on a public CETSA dataset.
• Successfully translated CETSA features across cell lines.
• Validated the model's ability to predict cross-cell line CETSA profiles.

Conclusions:

• CycleDNN offers an effective computational approach for CETSA feature translation.
• The model has the potential to accelerate drug discovery and biological studies.
• This method reduces the need for extensive experimental validation across multiple cell lines.