Validation ofde novodesigned water-soluble and transmembrane proteins byin silicofolding and melting

Abstract

In silico validation of de novo designed proteins with deep learning (DL)-based structure prediction algorithms has become a mainstream practice. However, formal evidence of the relationship between a high-quality predicted model and the chances of experimental success is lacking. We used experimentally characterized de novo designs to show that AlphaFold2 and ESMFold excel at different tasks. ESMFold can efficiently identify designs generated based on high-quality (designable) backbones. However, only AlphaFold2 can predict which sequences have the best chance of experimentally folding among similar designs. We show that ESMFold can generate high-quality structures from just a few predicted contacts and introduce a new approach based on incremental perturbation of the prediction (in silico melting), which can reveal differences between designs in terms of present favorable contacts. This study provides insight on how DL-based structure prediction could improve de novo protein design and open the door to the design of increasingly complex proteins; in particular membrane proteins which have historically been lacking basic in silico validation tools.

Sculpting conducting nanopore size and shape throughde novoprotein design

“…Following several iterations of combinatorial sequence design and structure relaxation, designs were selected based on hydrogen bond network descriptors, secondary structure ( 28 ) and aggregation propensities ( 29 ) (Figure S4). We previously found that AlphaFold2 ( 30 ) could accurately predict the structures of designed TMBs even in the absence of evolution information (from a single sequence input and without a multiple sequence alignment ( 31 )) when weak 3D contacts contained in the sequence were amplified by 48 rounds of molecular model recycling through the prediction network, and that the confidence assigned to the model (plDDT) was a good discriminator of the sequences with higher probability of experimentally folding ( 32 ). Therefore, we selected 4-10 designs per blueprint for which AlphaFold2 predicted high-confidence structures closely matching the design models (Figure S5).…”

“…ESM noise ramp ( in silico melting ( 21 ) simulations) plots for (A) square-shaped TMB12 designs and (B) rectangle-shaped designs (curves colored by protein expression level, as defined in Figure S6. red=strongly expressed; orange=weakly expressed; gray=no expression).…”