A convolutional neural network for the prediction and forward design of ribozyme-based gene-control elements

Area of Science:

• Molecular Biology
• Biotechnology
• Bioinformatics

Background:

• Ribozyme switches are RNA-based genetic switches controlling gene expression.
• Understanding sequence-structure-activity relationships is key for designing novel ribozymes.

Purpose of the Study:

• To develop predictive models for ribozyme switch activity using machine learning.
• To enable the de novo rational design of functional ribozyme switches.

Main Methods:

• Generated large datasets of ribozyme sequence activity.
• Employed automated structural analysis and machine learning algorithms.
• Trained predictive models to estimate in vivo gene-regulatory activity.

Main Results:

• Developed accurate predictive models for ribozyme gene-regulatory activity.
• Designed ribozyme libraries with low basal activity.
• Created novel ribozyme switches responsive to specific ligands, achieving high success rates.

Conclusions:

• Machine learning can accurately predict RNA gene-regulatory activity.
• This approach facilitates the rational design of functional RNA molecules.
• Provides a foundation for developing advanced RNA design tools.