Generative aptamer discovery using RaptGen.

Area of Science:

• Biotechnology
• Computational Biology
• Molecular Biology

Background:

• Nucleic acid aptamers are crucial molecular tools generated through systematic evolution of ligands by exponential enrichment (SELEX).
• Current SELEX methods are often constrained by the volume and scope of experimental sequencing data.
• There is a need for computational approaches to enhance aptamer discovery and design.

Purpose of the Study:

• To introduce RaptGen, a novel variational autoencoder model for in silico aptamer generation.
• To demonstrate RaptGen's capability in effectively representing and embedding aptamer sequence motifs.
• To validate RaptGen's utility in generating novel aptamers, including truncated sequences, and its applicability in activity-guided discovery.

Main Methods:

• Development of RaptGen, a variational autoencoder incorporating a profile hidden Markov model decoder.
• Embedding of simulated and experimental sequence data into a low-dimensional latent space.
• Application of Bayesian optimization for activity-guided aptamer generation.

Main Results:

• RaptGen successfully embedded sequence data into a latent space based on motif information.
• The model generated novel aptamers from the latent space, even those absent in high-throughput sequencing data.
• RaptGen demonstrated the ability to generate truncated aptamers and facilitate activity-guided aptamer discovery.

Conclusions:

• RaptGen offers a powerful generative approach for in silico aptamer discovery.
• Latent space representation is a valuable strategy for identifying and designing functional aptamers.
• The RaptGen method advances computational strategies in molecular evolution and aptamer design.