The number of viable ways to build a working protein is 1,000 trillion times larger than the narrow paths chosen by natural evolution.
Natural proteins only represent a tiny fraction of what is actually possible in biology. High-entropy generative models have discovered a vast ocean of functional sequences that nature has never tried. These artificial proteins work just as well as natural ones but look completely different at the sequence level. This finding suggests that the requirements for life-sustaining molecules are much more flexible than once believed. We can now look far beyond the family trees of known organisms to invent entirely new biological functions.
Expanding functional protein sequence space using high entropy generative models
arXiv · 2605.03578
Boltzmann Machines trained on evolutionary sequence data have emerged as a powerful paradigm for the data-driven design of artificial proteins. However, the relationship between model architecture, specifically parameter density, and experimental performance remains poorly understood. Here, we investigate this relationship using the Chorismate Mutase enzyme family as a model system. We compare standard fully connected Boltzmann Machines for Direct Coupling Analysis (bmDCA) with sparse models gen