Disruptive mixed in vitro-in silico approach for protein engineering and screening
September 24-28, 2017
We present a strategy that combines wet-lab experimentation and computational protein design for engineering polypeptide chains. The protein sequences were numerically coded and then processed using Fourier Transform (FT). Fourier coefficients were used to calculate the energy spectra called "protein spectrum". We use the protein spectrum to model the biological activity/fitness of protein from sequence data. We assume that the protein fitness (catalytic efficacy, thermostabilty, binding affinity, aggregation, stability…) is not purely local, but globally distributed over the linear sequence of the protein. Our patented method does not require any protein 3D structure information and find patterns that correlate with changes in protein activity (or fitness) upon amino acids residue substitutions. A minimal wet lab data sampled from mutation libraries (single or multiple points mutations) were used as learning data sets in heuristic approaches that were applied to build predictive models. We show the performance of the approach on designed libraries for 3 examples (enantioselectivity, thermostability and binding affinity) and discuss how our approach can tackle epistatic phenomena. We can screen up to 1 billion (109) protein variants in a very short time.
Frederic Cadet, Matthieu NG FUK CHONG, and Nicolas FONTAINE, "Disruptive mixed in vitro-in silico approach for protein engineering and screening" in "Enzyme Engineering XXIV", Pierre Monsan, Toulouse White Biotechnology, France Magali Remaud-Simeon, LISBP-INSA, University of Toulouse, France Eds, ECI Symposium Series, (2017). http://dc.engconfintl.org/enzyme_xxiv/87