Deep learning in biology faces a transferability crisis

Key Points

• The aim is to address the challenges of generalizability in deep learning, particularly in biosciences.
• Identified issues with transferability in deep learning models.
• Analyzed misleading claims surrounding model performance.
• Suggested frameworks for better evaluation of models in biological contexts.
• Outlined the extent of the transferability crisis in biosciences.
• Proposed actionable solutions to improve model evaluation and transferability.