Structure-enhanced deep learning accelerates aptamer selection for small molecule families like steroids.

The efficient discovery of high-affinity small-molecule aptamers via the Systematic Evolution of Ligands by EXponential enrichment (SELEX) is often constrained by challenges in navigating vast sequence spaces and rationally designing initial libraries. In this study, we introduce Deep Learning-assisted SELEX (DL-SELEX), a novel two-step framework that employs variational autoencoders (VAEs) to accelerate and refine small-molecule aptamer selection. This approach is the first to integrate deep learning to design initial aptamer libraries, marking a significant advancement in SELEX workflows. DL-SELEX leverages shared structural features within molecular families (e.g. steroids) to guide aptamer design: AptaVAE, the first VAE enriched with transfer learning from foundation models, generates tailored initial pools, whereas AptaClux, a second VAE, identifies high-performance candidates from SELEX-derived next-generation sequencing (NGS) data by capturing consensus structural features. The application of DL-SELEX to hydrocortisone (CS) and testosterone (TES) yielded aptamers with up to 450-fold higher affinity than previously reported aptamers and reduced SELEX iterations by up to 80%. Critically, these results demonstrate that structural commonalities can be used to train deep learning models to design aptamers for structurally similar targets. DL-SELEX provides an effective, generalizable strategy to streamline aptamer discovery and enables de novo design of high-affinity aptamers for challenging small molecules.