Proteins are natural polymers that play a crucial role both in living organisms and in biotechnological applications. In biological and industrial processes, they can be exposed to mechanical influences, such as shear flow. In this work, we aim to exploit the effect of shear flow on a protein to control its orientation. To simulate a protein solvated in water, we employ molecular dynamics simulations in combination with the Martini force field. Control of the protein orientation is carried out using reinforcement learning with the Q-learning method. The procedure consists of a cycle of a short simulation, determination of the new protein orientation, and selection of new shear stress parameters for the simulation in the next iteration of the cycle. The results of reinforcement learning, such as the trajectory of protein orientation, are compared with a reference physical prediction, which is based on the assumption that shear flow can induce rotation only around two fixed axes in space.
|
