Biocomputing platforms, such as cultured neurospheres, have the potential to provide great advances in biohybrid computation and control systems. However, to design and fabricate neurospheres reliably and reproducibly, models that can accurately predict their computational behaviors are required. Towards the end of understanding how neurospheres perform higher-level computations, we present a model framework for simulating the dynamics of stochastically-connected neuron networks. Each neuron is modeled using biophysical models of neural excitability, and the system can be stimulated by a small number of simulated electrodes. The network response of the system is analyzed using Principal Components Analysis on the firing frequencies of the neurons. From preliminary simulations, we demonstrate the ability of these neurosphere networks to encode information about the magnitude of current stimuli. Future additions to this framework will incorporate the 3D geometry of both the neurosphere and individual dendritic trees.