Low-Data Drug Design with Few-Shot Generative Domain Adaptation

Area of Science:

• Drug Discovery
• Artificial Intelligence
• Computational Chemistry

Background:

• Developing drugs for emerging diseases is critical for public health.
• Artificial intelligence (AI) accelerates drug discovery, particularly generative models like Generative Adversarial Networks (GANs).
• Limited data for new diseases challenges the effectiveness of standard generative models for drug molecule discovery.

Purpose of the Study:

• To address the challenge of generating high-quality and diverse drug molecules with limited data for new diseases.
• To propose a novel molecule generative domain adaptation paradigm (Mol-GenDA).
• To enable the transfer of knowledge from pre-trained GANs to new disease domains with minimal reference data.

Main Methods:

• Developed a molecule generative domain adaptation (Mol-GenDA) paradigm.
• Introduced a molecule adaptor into the GAN generator during fine-tuning.
• Transferred a pre-trained GAN from a large dataset to a new disease domain using few references.

Main Results:

• Mol-GenDA effectively generates high-quality and diverse drug candidates.
• The approach successfully reuses prior knowledge from pre-training.
• Maintains molecule quality and diversity under limited supervision.

Conclusions:

• Mol-GenDA offers a promising solution for expedited drug discovery in emerging diseases.
• The method addresses the low-data challenge in generative drug design.
• Facilitates timely development of effective drugs against new outbreaks.