Fig. 1: The workflow of this study.

a We prepared brain slices from individual regions, as illustrated here with orange circles. We used two of these regions for the training step and the generation step, respectively. b Prior to this, we classified the cortical regions into 16 groups. The abbreviated names of the 16 areas are defined by the following rules: Right and left hemispheres include 8 groups, respectively, and expressed as L or R at the beginnings of individual names. The name is followed by combinations of O, D, F, and V expressing abbreviations of Occipital, Dorsal, Frontal, and Ventral. (Refer to the supplemental material. 1 detailed locations of the slices used for the 16 area groups.) Pairs of regions, like the examples in a, were selected from those 16 groups. c We measured neuronal activity from hundreds of neurons in each region with a multi-electrode device. d An example of a spike train obtained from one of the measurements. The horizontal axis is time [sec] and the vertical axis is the index of neurons. The timing at which a certain neuron fires is indicated by a dot. This diagram is called a raster plot in neuroscience. e We used stained images to extract only neuron groups in cortical areas and then divided the neuron groups with lines orthogonal to the cortex so that only 128 cells were included in each dataset. f Such spike sequence, binary data, is input to the Multilayer LSTM model to predict one step ahead after learning from the past data. The horizontal axis is time [ms], and the input vector is a binary vector.