Machine learning guided design of strong transcriptional activation domains

More than 90% of 16,770 well-measured candidate activation domains designed via machine learning function in yeast, showing strong sequence-to-function predictability.

• Why it matters: Designing intrinsically disordered activation domains is hard because sequence constraints are weak and sequence-to-function links are poorly defined, limiting de novo functional design.
• What they did: The team generated 500,000 candidate AD sequences resembling known ones, scored them with three neural networks and other models, then experimentally tested 18,750 in yeast.
• The result: Consensus scoring across models is linked to stronger AD activity, enabling MADRIGAL and supporting designed ADs that retain function across fungi, plants, and human cells.

AI summary of the abstract · may contain errors · not medical advice · read the paper.