Ice-nucleation active (INA) bacteria can promote the growth of ice more effectively than any other known material. Using specialized ice-nucleating proteins (INPs), they obtain nutrients from plants by inducing frost damage and, when airborne in the atmosphere, they drive ice nucleation within clouds, which may affect global precipitation patterns. Despite their evident environmental importance, the molecular mechanisms behind INP-induced freezing have remained largely elusive. We investigate the structural basis for the interactions between water and the ice-nucleating protein InaZ from the INA bacterium Pseudomonas syringae. Using vibrational sum-frequency generation (SFG) and two-dimensional infrared spectroscopy, we demonstrate that the ice-active repeats of InaZ adopt a β-helical structure in solution and at water surfaces. In this configuration, interaction between INPs and water molecules imposes structural ordering on the adjacent water network. The observed order of water increases as the interface is cooled to temperatures close to the melting point of water. Experimental SFG data combined with molecular-dynamics simulations and spectral calculations show that InaZ reorients at lower temperatures. This reorientation can enhance water interactions, and thereby the effectiveness of ice nucleation.
Attention mechanisms have led to many breakthroughs in sequential data modeling but have yet to be incorporated into any generative algorithms for molecular design. Here we explore the impact of adding self-attention layers to generative β-VAE models and show that those with attention are able to learn a complex “molecular grammar” while improving performance on downstream tasks such as accurately sampling from the latent space (“model memory”) or exploring novel chemistries not present in the training data. There is a notable relationship between a model's architecture, the structure of its latent memory and its performance during inference. We demonstrate that there is an unavoidable tradeoff between model exploration and validity that is a function of the complexity of the latent memory. However, novel sampling schemes may be used that optimize this tradeoff. We anticipate that attention will play an important role in future molecular design algorithms that can make efficient use of the detailed molecular substructures learned by the transformer.