Seminars – Center for Microbial Dynamics and Infection

Upcoming schedule here. This trainee-run seminar series focused on new research in CMDI. It’s Fridays at 3pm and includes coffee and cookies. Spring 2026 seminars will be held in Cherry Emerson room 320.

Upcoming seminar:

Abstract:

Evolution is a dynamic process, with environments constantly shifting. Every living organism is the endpoint of a long history, only surviving by moving from environment to environment as opportunities present themselves. As they evolve in a given environment, organisms gain adaptations – but what happens to those novel adaptations when they move on? The actual past history of living organisms is inaccessible, and most experimental evolution studying multiple environments has studied rapidly fluctuating environments which drive the evolution of versatility rather than serial adaptation to novelties. Using long term evolution experiments, I demonstrate that when yeast are transferred from one environment to another they have not been exposed to previously, recent adaptations are lost extremely rapidly – far more rapidly than adaptations are gained to the new environment. However, I also demonstrate that this is not due to intrinsic tradeoffs between performance in multiple environments. Instead, this appears to be due to the phenomenon of global epistasis, by which as fitness increases in an environment helpful mutations become scarce while harmful mutations become plentiful. I present a model of epistatic interactions in multiple environments, in which as evolution switches from one environment to another pleiotropy and epistatic interactions can rapidly destroy recent adaptations while having little effect on fitness in environments seen less recently. This phenomenon has striking and surprising similarities to the phenomenon of “catastrophic forgetting” from machine learning, by which a neural network switched between training tasks forgets performance on past tasks in favor of the most recent task. This suggests a remarkable connection between phenomena observed in biological evolution and machine learning, with possible deep mathematical and phenomenological connections between the two fields.