Diversifying Design of Nucleic Acid Aptamers Using Unsupervised Machine Learning.

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Aptamer design traditionally relies on empirical methods like SELEX, often leading to limited sequence diversity.
• Discovering novel aptamers with diverse chemical compositions is crucial for expanding their therapeutic and diagnostic potential.
• Systematic protocols for generating diverse aptamer candidates are needed to overcome limitations of current design strategies.

Purpose of the Study:

• To develop a machine learning-based method for the inverse design of single-stranded RNA and DNA aptamers.
• To enable the generation of novel aptamer sequences with controllable sequence diversity.
• To facilitate the discovery of useful nucleic acid aptamers with desired properties and distinct chemical compositions.

Main Methods:

• Utilized an unsupervised machine learning model, the Potts model, trained on a small set of empirically identified sequences using the maximum entropy principle.
• Leveraged the spectral feature (energy bandgap) of the Potts model to control sequence diversity.
• Sampled sequences within a specific Potts energy range to generate distinct yet functionally relevant candidates.

Main Results:

• Successfully applied the Potts model to design diverse pools of 30-mer RNA and DNA aptamers.
• Demonstrated the ability to generate sequences with specified secondary structure motifs.
• Showcased controllable sequence diversity in the generated aptamer candidates.

Conclusions:

• The proposed Potts model approach offers a systematic and effective strategy for aptamer inverse design.
• This method enhances the discovery of novel nucleic acid aptamers by enabling controllable sequence diversity.
• The approach holds promise for generating diverse aptamer libraries with specific functional features for various applications.