Methods and systems for protein engineering and production

The method of any preceding claim, wherein constant parts are up to about 2000 nucleotide long, and/or wherein variable parts are up to about 200 nucleotide long.

The method of any preceding claim, wherein each sequence variant comprises a plurality of constant parts and/or a plurality of variable parts.

The method of any preceding claim, wherein the library design step (a) comprises designing at least one of the one or more variable regions to include random variability in at least one position, optionally wherein the library design step (a) comprises designing at least one of the one or more variable regions to include random variability in one or more specific positions of the at least one variable region.

The method of any preceding claim, wherein at least one of the one or more constant regions comprises one or more sequences selected from:

• a promoter sequence, an enhancer sequence, a localisation signal, a flag sequence, a marker sequence, a ribosome binding site, a stop codon, a start codon, a 5′ stem loop structure, a 3′ stem loop culture, an origin of replication and a selection sequence.

The method of any preceding claim, further comprising a step (a″) of producing the proteins encoded by each sequence variant of the nucleic acid library to obtain a protein library, wherein the library testing step (b) comprises subjecting the protein library to one or more assays to test for the one or more desired properties.

The method of any preceding claim, wherein the one or more desired properties is/are chosen from:

• physico-chemical properties of the proteins, activity-related properties, physiologically-relevant properties, and pharmacokinetic properties.

The method of any preceding claim, wherein the library testing step comprises testing the variants for a plurality of properties, and the learning step comprises assigning a plurality of fitness scores to each variant tested, wherein each fitness scores corresponds to one of the plurality of properties, wherein the learning step comprises training a plurality of machine learning algorithms, wherein each machine learning algorithm is trained to predict at least one of the plurality of fitness scores for new sequence variants.

The method of any preceding claim, wherein the machine learning algorithm is a classifier, wherein the machine learning algorithm is a neural network.

The method of any preceding claim, wherein the machine learning model trained in step (c) is used to design a new library of sequence variants by iteratively optimising a library of sequence variants in silico, optionally wherein the library of sequence variants is iteratively optimised using a genetic algorithm.

The method of any preceding claim, further comprising repeating steps (a) to (c) with the new library.

The method of any preceding claim, wherein the new library comprises at least one sequence variant encoding for a protein with the one or more desired properties.

The method of any preceding claim, wherein the new library of sequence variants with an improved distribution of fitness scores is one wherein at least 30% of the sequence variants have one or more variable regions having a DNA sequence similarity of less than 95% with respect to the corresponding one or more variable regions of all, or a proportion of, the sequence variants within the library prepared in step (a).

The method of any preceding claim, wherein a higher proportion of sequence variants of the new library display one or more improved desirable properties compared to the sequence variants within the library prepared in step (a).