EXPERIMENTS IN THE USE OF ΔΔ𝐺 PREDICTIONS FOR DENOVO IN-SILICO PROTEIN DESIGN

Abstract

"The biosphere solution to the protein design problem is akin to a planetwide computational machine running a simple stochastic algorithm. The root of its success lies in the open endedness and diversity of its search, willing to extend across all possible niches over time. Nature’s work produces living transient DNA artifacts that are stored in the biosphere, and, at an increasing rate, being read by sequencing technology and transfered to large datasets now hosting millions of raw samples. This data can be leveraged by the use of increasingly large pLM(protein Language models), that have been shown to essentially function as unsupervised protein structure learners. (...)"