view article
Figure 3
A schematic representation of the analysis method for TRX measurements using KINNTREX. The NN consists of two sub-networks. The first network is called the projection NN. It aims to predict the input, EM (M for measured), as accurately as possible by generating time-dependent DED maps (EC1, C for calculated) from significant lSVs (U) along with the DED maps of the intermediates (I) and the concentrations (CNN). The second sub-network called conversion NN takes the CNN as input and predicts RRCs, k. CNN is flattened before being applied to the conversion NN. After passing through both sub-networks, KINNTREX solves the differential equation governing the kinetic mechanism of the protein photocycle (red dashed box), resulting in the concentrations CCDE. In a subsequent step, prior to the calculation of the loss function, the time-dependent DED maps, EC2, are predicted a second time using the DED maps of the intermediates I from the projection NN and the CCDE [equation (4[link])]. The loss function (purple dashed box) evaluates the discrepancies between measured and predicted time-dependent DED maps as well as the differences between the calculated concentrations (CNN and CCDE). The user can constrain the adjustable range of the RRCs to further inform the loss function. A backpropagation procedure concludes the NN. The arrows form a loop that iterates multiple times.

IUCrJ
Volume 11| Part 3| May 2024| Pages 405-422
ISSN: 2052-2525
https://doi.org/10.1107/S2052252524002392