AI-Guided Resetting of Memories in Gene Regulatory Network Models: biomedical and evolutionary implications

Abstract

Molecular pathways such as gene-regulatory networks regulate numerous functions in cells and tissues that impact embryonic development, regenerative repair, aging, cancer, and many other aspects of health and disease. One important aspect of such networks is experience-dependent plasticity: their activity changes after repeated exposures to external and internal physiological stimuli. This kind of functional plasticity gives rise to habituation to pharmacological interventions (resulting in loss of efficacy over time), sensitization (resulting in unacceptable side effects after repeated use), or canalizing of undesired dynamics that recur even when the original problem has been resolved (persistent disease states). Our in silico analyses reveal that memories formed in gene regulatory networks can be erased by specific further experiences without any changes of network topology (leaving the connectivity in place). We present a method for discovery of stimuli that can be used to selectively delete physiological memories, which can be used to remove unwanted behaviors in biomedical and bioengineering contexts without gene therapy or genomic editing. Remarkably, not only are the training-induced gains in causal emergence not lost after stimuli that wipe memories, but we also find a positive relationship between the causal emergence and learning ability of a network, suggesting a deep asymmetry (ratchet) in the relationship between learning/forgetting and integration of collective intelligence which may have implications for evolution. Significance StatementWe present an AI-driven method for discovering signals that can be used to induce physiological networks to forget specific behaviors, which can be used for applications in biomedicine and bioengineering.