TransferBAN-Syn: a transfer learning-based algorithm for predicting synergistic drug combinations against...

Area of Science:

• Parasitology
• Computational Biology
• Drug Discovery

Background:

• Echinococcosis is a significant zoonotic parasitic disease requiring effective treatments.
• Drug combination therapy is crucial for overcoming drug resistance and enhancing efficacy in echinococcosis.
• Traditional experimental methods for identifying drug combinations are inefficient and costly, with limited data available for echinococcosis.

Purpose of the Study:

• To develop a computational model for identifying synergistic drug combinations against echinococcocosis.
• To address the challenge of limited drug combination data in echinococcosis by employing transfer learning.
• To provide a novel computational approach for predicting drug pairs for diseases with scarce data.

Main Methods:

• A transfer learning-based model, TransferBAN-Syn, was developed.
• The model utilizes a bilinear attention network for deep feature extraction of drug interactions.
• TransferBAN-Syn was trained on a dataset of 21 parasitic diseases (source domain) and fine-tuned for echinococcosis (target domain).

Main Results:

• TransferBAN-Syn significantly improved the accuracy of predicting synergistic drug combinations for echinococcosis.
• The model demonstrated enhanced generalizability compared to traditional methods.
• The study identified promising novel drug combinations for echinococcosis treatment.

Conclusions:

• TransferBAN-Syn offers a powerful and accurate computational approach for discovering synergistic drug combinations in echinococcosis.
• This transfer learning strategy effectively overcomes data limitations in drug discovery for rare or understudied diseases.
• The model provides a valuable tool for advancing echinococcosis treatment and offers a blueprint for similar challenges in other diseases.