Accelerating Aptamer Discovery using a Transformer Based T-5 Model: A Preliminary Study

Aptamers are short nucleic acid sequences that bind targets with high specificity,but their discovery via SELEX is time-consuming and biased. We propose atransformer-based T5 model that is fine-tuned for aptamer generation, incorpo-rating primary sequence and secondary structural information to minimize thetime required in small molecule aptamer SELEX. This preliminary study aimsto identify high-affinity target sequences as early as possible in the selection pro-cess, reducing the number of required experimental rounds. The model generateshigh-affinity sequences, achieving strong alignment scores with the top 0.1% ofexperimentally validated aptamers, and demonstrates cross-target aptamer gener-ation (with top 10% existing aptamers), enabling discovery for structurally sim-ilar molecules. Our custom RAG and fine-tuning prioritizes the most promisingsequences for experimental validation, streamlining the selection process and ac-celerating experimental workflows