Modelling linear spectra of plant light-harvesting complexes

This datasets consist of four Keras neural network (ANN) models and simulated linear spectra.  

Three of the models predict the pigment energies and other molecular parameters of the light-harvesting complex CP29 from linear spectra at 77 K and room temperature. The spectra required are the absorption, fluorescence, linear dichroism, and circular dichroism (CD) spectra. The ANNs take as input a joined vector of the spectra and they output the molecular parameters. One of the models predict all parameters from all of the spectra. Another uses all spectra apart from CD spectra to predict all of the parameters. The third model predicts only the energies (i.e., not the other molecular parameters) from all of the spectra.  It should be clear from the filenames which is which. The fourth model predicts the linear spectra from the energies and molecular parameters. 

The spectral data sets contain absorption- and fluorescence-type linear spectra calculated for a dimer system with a system-bath interaction similar to that of plant light-harvesting complexes. A phonon continuum and eight chlorophyl high-frequency modes were included in the spectral density that characterises the system-bath interaction. Spectra were calculated for different site energy gaps (0 to 500 per cm) and couplings (-55 to 55 per cm) and also for different exciton energy gaps for strong couplings (>300 per cm). To calculate the spectra, an exact stochastic path integral method was used, together with a number of approximate methods (the Full Cumulant Expansion, complex time-dependent Redfield (ctR), Redfield, modified Redfield, and a reorganisation-shifted version of the ctR method---the latter is applicable only for fluorescence spectra). The energy gaps, couplings, and dipole factors are indicated in the files.