Machine learning-powered, high-affinity modification strategies for aptamers

Abstract

The binding affinity of aptamers to targets has a crucial role in the pharmaceutical and biosensing effects. Despite diverse post-systematic evolution of ligands by exponential enrichment (post-SELEX) modifications explored in aptamer optimization, accurate prediction of high-affinity modification strategies remains challenging. Sclerostin, which antagonizes the Wnt signaling pathway, negatively regulates bone formation. Our screened sclerostin aptamer was previously shown to exert bone anabolic potential. In the current study, an interactive methodology involving the exchange of mutual information between experimental endeavors and machine learning was initially proposed to design a high-affinity post-SELEX modification strategy for aptamers. After four rounds of interactive training (a total of 422 modified aptamer-target affinity datasets with diverse modification types and sites), an antifcial intelligence model with high predictive accuracy with a correlation coefficient of 0.82 between the predicted and actual binding affinities was obtained. Notably, the machine learning-powered modified aptamer selected from this work exhibited 105-fold higher affinity (picomole level K D value) and a 3.2-folds greater Wnt-signal re-activation effect compared to naturally unmodified aptamers. This approach harnessed the power of machine learning to predict the most promising high-affinity modification strategy for aptamers.