Rapid evolution follows predictable paths on learnable low-dimensional manifolds

In this study, we explore the potential of in-silico evolution to unravel the complexities of gene regulatory networks (GRNs) in responding to diverse input patterns. Utilizing computational techniques, we evolve GRNs to optimize their fitness profiles across varying environmental inputs. Subsequently, we employ a geometric variational autoencoder (VAE) to train on these fitness profiles, aiming to identify low-dimensional representations of the evolutionary trajectories. This approach allows us to capture the underlying dynamics and constraints governing GRN evolution. To validate our findings, we compare the inferred low-dimensional representations with experimental data on rapid adaptation observed in antibiotic resistance scenarios. By integrating computational evolution with advanced machine learning techniques, our work provides novel insights into the fundamental mechanisms of rapid adaptation, highlighting the power of combining in-silico models with experimental data to enhance our understanding of evolutionary processes and their applications in combating antibiotic resistance.