Designing molecular RNA switches with Restricted Boltzmann machines.

Area of Science:

• RNA biology
• Computational biology
• Biochemistry

Background:

• Riboswitches are allosteric RNA molecules that regulate gene expression by changing conformation upon metabolite binding.
• Aptamers are the core components of riboswitches responsible for metabolite recognition and conformational changes.
• De novo design of functional RNA molecules remains a significant challenge in synthetic biology.

Purpose of the Study:

• To develop a machine learning approach for the de novo design of riboswitch-like aptamers.
• To generate novel SAM-I riboswitch aptamers with desired allosteric functions.
• To experimentally validate the conformational switching capabilities of designed aptamers.

Main Methods:

• Utilized Restricted Boltzmann Machines (RBM) to learn generative models from homologous RNA sequences.
• Applied RBM to design new SAM-I riboswitch aptamers.
• Employed chemical probing techniques (SHAPE and DMS) for high-throughput experimental validation.
• Developed a tailored analysis pipeline for assessing conformational changes in designed sequences.

Main Results:

• RBM models accurately captured sequence conservation, covariation, and diversity of natural aptamers across four families.
• 476 de novo designed SAM-I aptamers were experimentally probed alongside 201 natural sequences.
• Designed aptamers with high RBM scores and 20-40% sequence divergence showed a ~30% success rate in SAM-induced conformational switching.
• The ability of designed aptamers to switch conformation correlated with specific energetic features.

Conclusions:

• Restricted Boltzmann Machines are effective for the de novo design of functional riboswitch aptamers.
• Designed aptamers can mimic the allosteric regulatory mechanism of natural riboswitches.
• The study provides insights into the energetic determinants of RNA conformational switching, paving the way for designing novel RNA-based regulatory elements.