Machine learning guided aptamer refinement and discovery.

Area of Science:

• Biotechnology
• Computational Biology
• Nucleic Acid Chemistry

Background:

• Aptamers are nucleic acid ligands with high affinity and specificity for target molecules.
• Discovering optimal aptamers involves searching vast sequence spaces, which is practically limited.
• Machine learning offers a way to intelligently navigate sequence space for aptamer identification.

Purpose of the Study:

• To develop a machine learning approach combined with particle display for efficient aptamer discovery.
• To train in silico models for predicting aptamer affinity using experimental data.
• To accelerate the identification and design of high-performance aptamers for therapeutic and diagnostic applications.

Main Methods:

• Utilized particle display (PD) to partition aptamer libraries based on binding affinity.
• Trained machine learning models using PD-generated affinity data to predict aptamer performance.
• Employed the trained models to predict and generate novel high-affinity aptamers.

Main Results:

• The machine learning model predicted high-affinity DNA aptamers with an 11-fold improvement over random methods.
• Novel, high-affinity aptamers were generated at a higher rate than with PD alone.
• Designed truncated aptamers that were 70% shorter and exhibited superior binding affinity (1.5 nM).

Conclusions:

• Combining machine learning with physical methods like particle display significantly enhances aptamer discovery efficiency.
• This integrated approach can expedite the development of improved diagnostic and therapeutic agents.
• The strategy enables the design of optimized aptamers with enhanced properties, such as reduced size and increased affinity.