Simulating Cytotoxic T-Lymphocyte and Cancer Cells Interactions - An LSTM-Based Approach to Surrogate an Agent-Based Model.

Through an Agent-based model (ABM) it is possible to compute simple and basic behaviours at the cellular scale, while observing the emergence of complex conducts or patterns at the population scale. Thus, in this modeling paradigm, macroscopic phenomena can be explained by a set of behaviors of the agents. However, due to the high computational cost, the exploration of the parameters of these models for the optimization or calibration of protocols is still an open challenge. In this paper, we propose a surrogate model based on an Long-Short Term Memory (LSTM) neural network to replicate the predictions of an ABM much faster. The ABM used in this paper models the interactions between cytotoxic T-lymphocytes (CTL) and cancer cells [ 8 ]. The initial results shows that the neural network is capable of reproducing the emergent behavior of the ABM with a reduced computational cost.