Combining AlphaFold 3 with microscale thermophoresis for aptamer optimisation

In the microscopic world of biology, scientists often need to find a specific key that fits a specific lock. Sometimes, the key is a protein, a complex machine built by a cell, and the lock is a disease-causing molecule. Other times, the roles are reversed, and researchers design a tiny piece of genetic material, called an aptamer, to act as the key. These aptamers are like custom-made molecular hooks, engineered to grab onto a target with incredible precision. Once they find their match, they can stop a harmful process, like a bacterium building a sticky film known as a biofilm that protects it from medicine. For years, figuring out exactly how these hooks and locks fit together required expensive, slow, and difficult experiments. Recently, a powerful new computer program called AlphaFold 3 arrived, promising to predict these 3D shapes instantly, as if the computer could simply imagine the solution before the scientists ever mixed chemicals in a lab.

However, a computer prediction is only as good as the reality it tries to describe. In a recent study, researchers set out to test whether this new digital crystal ball could truly replace the messy, physical work of the laboratory. They focused on a specific aptamer designed to stop a common bacterium, Streptococcus mutans, from forming biofilms on teeth. This bacterium produces a protein called GbpC, which acts like a glue, helping the bacteria stick together. The researchers had already found a genetic key, named A39, that could block this glue. But they did not know exactly where on the protein the key touched or how the two molecules held hands. To solve this, they used the computer to build a model of the interaction and then tested that model against real-world measurements to see if the digital guess matched the physical truth.

The team started by feeding the computer the sequence of the aptamer and the protein. The software, AlphaFold 3, generated a 3D model showing how the two might fit together. The program suggested that the aptamer did not use its entire length to bind to the protein. Instead, it predicted that a very short, specific stretch of genetic letters formed a tight grip inside a groove on the protein, right next to a calcium ion. Based on this digital blueprint, the researchers designed five new versions of the aptamer. They created one that kept only the predicted gripping part, another that removed the parts the computer said were useless, and others that changed the structure to see if the grip would hold. They then ran these new designs through a computer simulation to see how confident the program was in its own predictions. The computer gave high scores to models where the specific gripping sequence was present, suggesting these were the most likely to work.

Next, the researchers moved from the screen to the bench to see if the computer was right. They mixed the real protein with the real aptamers in a liquid solution and used a technique called microscale thermophoresis. This method works by heating a tiny spot in the liquid; when the aptamer and protein stick together, they move differently in the heat than when they are apart. By watching this movement, the scientists could measure exactly how tightly the molecules held on to each other. The results were a mix of confirmation and surprise. The computer had correctly identified that a specific short sequence was essential for binding; when the researchers removed this sequence, the aptamer failed to stick to the protein at all. This confirmed that the computer could spot the general location of the interaction.

However, the computer failed to predict the strength of the bond. The program had suggested that a version of the aptamer containing only the short gripping sequence would be a strong binder, but in the real experiment, it did not bind at all. Conversely, the computer had predicted that a version with a shortened primer region (A39T) would still bind, but the real-world test showed that this version was significantly weaker than the original. Furthermore, the computer had predicted that a version lacking the main binding motif (A39CN) would bind via a different sequence, yet this variant showed no measurable binding in the lab. The most telling finding was that the computer's confidence scores did not match the actual strength of the bond. A model that the computer rated as highly reliable turned out to be a weak binder in the lab, while a model it rated with lower confidence actually held on quite well. The researchers found that while the computer could suggest where the molecules might touch, it could not tell them how hard they would hold on.

This study highlights a crucial lesson for the future of molecular design. While tools like AlphaFold 3 are powerful enough to generate hypotheses and point scientists toward the right area of a molecule, they are not yet ready to replace the need for physical testing. The computer can draw a map, but it cannot yet tell you how strong the terrain is. The researchers concluded that the best approach is to use these digital tools to guide experiments, but to always verify the results with real measurements. As more data from real experiments is fed back into these systems, the predictions may become more accurate, but for now, the laboratory bench remains the final judge of whether a molecular key actually fits the lock.