Joint CMU-Pitt Ph.D. Program in Computational Biology Seminar - Arvind Murugan

About this event

Cellular behavior is encoded in molecular interaction networks, which usually change only over generations, through evolution. Brains instead learn within a lifetime by tuning synapses through experience. We ask whether molecular systems can also learn from experience, without genetic change.

Such learning requires two ingredients: (a) tunable parameters and (b) a rule that tunes them. We argue that both arise naturally in many molecular systems, often through processes already known in other contexts. Molecular abundances provide tunable parameters in dense networks with mediated interactions, and simple processes such as binding-dependent degradation serve as the learning rule. Together, they implement a molecular form of Hebbian learning, in which molecules that spend time together grow to interact more strongly. Using theory and experiments, we show that such molecular mixtures can display Pavlovian conditioning, learn to perform classification, learn probabilistic behavior such as bet hedging, and learn to form condensates that reflect past environments. These results suggest that synthetic circuits could be trained rather than designed, and raise the question of how much non-genetic adaptation arises from ordinary molecular processes we do not recognize as learning.

Faculty Host: Christian Cuba Samaniego

In Person and Zoom: https://pitt.zoom.us/j/97970563600 Participation. See announcement.