Featuring the Bio Design Track Winners at the re:Agent Hackathon:
Meet AptaSense. Huge congrats to the team: Christopher Brenden,
Chengyong Yang, and Atul C..
Continuous monitoring changed diabetes care by replacing a number with a curve, but glucose is still nearly the only molecule we can watch that way: most of human chemistry is still measured one blood draw at a time. One way to make progress on this is to widen the range of targets that can be continuously monitored via electrochemical aptamer-based biosensors (EABs). But how do you turn an aptamer that binds its target into a sensor that also produces a strong, reliable signal?
Team AptaSense took this on, building a literature-, physics-, and data-informed pipeline for designing structure-switching EABs. Paperclip supplied the starting material. It searched the full text of the corpus for candidate aptamer sequences and read the surrounding text at each hit, so every candidate arrived with its citation and its context attached. Then the modeling ran on Proto, which gave the team a single place to reach for several tools at once. ViennaRNA did the physics, calculating folding energies for the clamped and open states of every candidate. That is what decides whether a design sits in the roughly 2 kcal/mol window where a switch actually switches. ESMFold2, AlphaFold3, Boltz-2 and Protenix co-folded each candidate against its target and against closely related proteins, so the pipeline could ask where a design lands and whether the models agree with each other. Every design came out the far end with the same set of numbers attached, which is what makes thousands of them rankable instead of a pile of sequences.
Since the relationship between aptamer sequence, structure, binding, and sensor signal is still difficult to predict reliably, they placed this pipeline within a lab-in-the-loop workflow, where high-throughput experiments generate data to improve future designs. That showed up directly in how the 96 wells got spent. No paper reports where IL-6 actually contacts this aptamer, so rather than filling a plate with minor variants of one guess, the agent swept four possible contact sites and discarded the two that yielded nothing usable, then tiled the survivors across eight energy bands — a plate that returns which designs switch and which structural hypothesis was right, from a synthesis run that was happening anyway. The long-term goal is to make EAB development faster and more reliable, helping expand continuous biosensing beyond glucose to a much broader range of molecules relevant to human health.
