METHODS, SYSTEMS, AND SOFTWARE FOR IDENTIFYING FUNCTIONAL BIO-MOLECULES

• US 20150065357A1
• Filed: 11/07/2014
• Published: 03/05/2015
• Est. Priority Date: 03/01/2002
• Status: Active Grant

• Current Assignee: Codexis, Inc.
• Original Assignee: Codexis Mayflower Holdings, LLC (Codexis, Inc.)
• Inventors: Fox, Richard John
• Granted Patent: US 10,453,554 B2
• Time in Patent Office: Days
• US Class Current: 506/2
• CPC Class Codes: C12N 15/1058 Directional evolution of libraries, e.g. evolution of libraries is achieved by mutagenesis and screening or selection of mixed population of organisms
• CPC Class Codes: G01N 33/6818 Sequencing of polypeptides
• CPC Class Codes: G06F 30/00 Computer-aided design [CAD]
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G16B 20/00 ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
• CPC Class Codes: G16B 20/20 Allele or variant detection, e.g. single nucleotide polymorphism [SNP] detection
• CPC Class Codes: G16B 20/50 Mutagenesis
• CPC Class Codes: G16B 30/00 ICT specially adapted for sequence analysis involving nucleotides or amino acids
• CPC Class Codes: G16B 30/10 Sequence alignment; Homology search
• CPC Class Codes: G16B 35/00 ICT specially adapted for in silico combinatorial libraries of nucleic acids, proteins or peptides
• CPC Class Codes: G16B 35/20 Screening of libraries
• CPC Class Codes: G16B 40/00 ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
• CPC Class Codes: G16B 40/20 Supervised data analysis
• CPC Class Codes: G16C 20/60 In silico combinatorial chemistry

Abstract

The present invention generally relates to methods of rapidly and efficiently searching biologically-related data space. More specifically, the invention includes methods of identifying bio-molecules with desired properties, or which are most suitable for acquiring such properties, from complex bio-molecule libraries or sets of such libraries. The invention also provides methods of modeling sequence-activity relationships. As many of the methods are computer-implemented, the invention additionally provides digital systems and software for performing these methods.

Claims

1-20. -20. (canceled)

21. A method for identifying amino acid residues for variation in a protein variant library in order to affect an activity of interest, the method comprising:

(a) receiving, for each protein variant in a training set, an amino acid sequence and the activity of interest obtained from assaying the protein variant;

(b) selecting amino acid residues and sequence positions of mutations in the training set;

(c) performing regression on the selected amino acid residues, sequence positions, and activities of the training set to produce a sequence-activity model for predicting the activity of interest as a function of multiple independent variables, the sequence-activity model comprising a plurality of linear terms and one or more non-linear terms,

wherein, for each non-linear term,

the non-linear term comprises a coefficient and two or more dummy independent variables,

the coefficient indicates the contribution to the activity of interest by the interaction of the two or more dummy independent variables, and

each of the two or more dummy independent variables specifies the presence or absence of a particular residue at a specific sequence position; and

(d) using the sequence-activity model to identify one or more amino acid residues at specific positions for varying or fixing to impact the activity of interest,

wherein operations (a)-(d) are performed by executing instructions on a computer system programmed to perform said operations.

22. The method of claim 21, wherein the non-linear term comprises a product of multiple factors, each factor comprising a different member of a group consisting of the coefficient and the two or more dummy independent variables.

23. The method of claim 22, wherein the non-linear term comprises a product of the coefficient and the two or more dummy independent variables

24. The method of claim 21, wherein the sequence-activity model does not include terms for at least some of the residues that do not vary in the protein variant library training set.

25. The method of claim 21, further comprising using the sequence-activity model to predict an activity of a new protein sequence.

26. The method of claim 21, wherein the regression performed in (c) comprises a partial least squares (PLS) regression.

27. The method of claim 21, wherein developing said sequence-activity model comprises selecting one or more non-linear terms from a group of potential non-linear terms.

28. The method of claim 27, wherein selecting the one or more non-linear terms comprises running a genetic algorithm to optimize one or more non-linear terms based upon the predictive ability of various models employing different non-linear terms.

29. The method of claim 28, wherein the predictive ability is measured by a fitness function of the genetic algorithm indicating how well the various models perform in predicting activity of proteins.

30. The method of claim 29, wherein the fitness function is applied to a validation set to perform a cross-validation, wherein the validation set comprises protein variants not belonging to the training set.

31. The method of claim 29, wherein the fitness function is measured by predicted residual sum of squares (PRESS), wherein the PRESS indicates the inaccuracy of a model in predicting activities of a plurality of protein variants.

32. The method of claim 29, wherein the genetic algorithm employs a convergence criterion that stops optimization when the fitness function of the best model from a number of successive generations of the genetic algorithm is identical or does not change appreciably.

33. The method of claim 28, wherein a current generation of models in the genetic algorithm comprise models obtained by mating parent models from a previous generation of models.

34. The method of claim 33, wherein the parent models from the previous generation comprise elite models in the previous generation, elite models being the fittest models in the previous generation.

35. The method of claim 33, wherein mating parent models from a previous generation of models comprises performing crossover by selecting one or more non-linear terms from a first parent model and one or more non-linear terms from a second parent model.

36. The method of claim 21, wherein using the sequence activity model to identify one or more amino acid residues comprises using the sequence activity model to rank residue positions in order of impact on the desired activity.

37. The method of claim 21, further comprising obtaining a sensitivity score of a mutated residue, wherein the sensitivity score indicates a relative influence of the mutated residue on the desired activity.

38. The method of claim 37, wherein the sensitivity score comprises a coefficient of a term.

39. The method of claim 37, wherein the sensitivity score indicates the average difference in activity between sequences having the mutated residue and corresponding sequences not having the mutated residue.

40. The method of claim 21, wherein the sequence activity model is used to identify one or more beneficial, neutral or detrimental mutations.

41. The method of claim 21, wherein the sequence activity model is used to identify one or more amino acid residues that are to remain fixed in a new protein variant library.

42. The method of claim 21, wherein using the sequence activity model comprises identifying a sequence predicted by the model to have a highest value of the activity of interest.

43. The method of claim 21, further comprising synthesizing one or more sequences having the one or more amino acid residues identified in (d).

44. The method of claim 43, further comprising assaying the synthesized sequences to determine the best-performing variant in vitro or in vivo.

45. The method of claim 43, further comprising using the synthesized sequences to produce a new protein variant library, thereby providing data to be received by operation (a) of claim 21 in a new round of directed evolution.

46. The method of claim 21, further comprising:

(e) generating a new protein variant library containing one or more new protein variants having amino acid sequences encoded by nucleic acids in which the identified nucleotides are varied or fixed in order to affect the activity of interest;

(f) assaying the new protein variant library to provide activity information for members of the new protein variant library; and

(g) producing a new sequence activity model from sequence and activity information for the members of the new protein variant library.

47. A method for determining the sequence of a protein having an activity of interest, the method comprising:

(a) providing, from a client computer to a computer network, an amino acid sequence and a measurement of the activity of interest for each of a plurality of training protein variants in a training set; and

(b) receiving, at the client computer from the computer network, a residue type and a sequence position for each of one or more amino acids in at least one selected protein variant, wherein the selected protein variant was identified by a sequence-activity model to have a particular level of the activity of interest,

wherein

the sequence-activity model comprises a plurality of linear terms and one or more non-linear terms,

the linear and non-linear terms are separated by plus or minus signs,

each linear term comprises the product of a coefficient and a bit-value independent variable, wherein the coefficient of the linear term indicates the relative impact on activity by the bit-value independent variable, and wherein the bit-value independent variable specifies the presence or absence of only one particular amino acid residue of a specific residue type at a specific sequence position, and

each non-linear term is a cross-product term comprising the product of a coefficient and two or more bit-value independent variables, wherein the coefficient of the non-linear term indicates the relative impact on activity by the interaction of the two or more bit-value independent variables, and wherein each of the two or more bit-value independent variables specifies the presence or absence of a particular residue of a specific residue type at a specific sequence position.

48. The method of claim 47, wherein the computer network is the Internet.

49. The method of claim 47, wherein obtaining the amino acid sequence and the measurement of the activity of interest comprises:

presenting, in a web browser on the client computer, a user interface that allows a user to upload a file containing the amino acid sequence and the measurement of the activity of interest; and

obtaining the amino acid sequence and the measurement of the activity of interest through the web browser.

50. A system for performing directed evolution of a protein variant library in order to affect an activity of interest, the system comprising:

one or more memory devices configured to store sequence and activity data; and

control logic configured to;

(a) receiving data characterizing a training set of a protein variant library, wherein the data provides activity and an amino acid sequence for each protein variant in the training set;

(b) from the received data, develop a sequence-activity model for predicting activity as a function of multiple independent variables, wherein

the sequence activity model comprises a plurality of linear terms and one or more non-linear terms,

the linear and non-linear terms are separated by plus or minus signs,

each linear term comprises the product of a coefficient and a bit-value independent variable, wherein the coefficient of the linear term indicates the relative impact on activity by the bit-value independent variable, and wherein the bit-value independent variable specifies the presence or absence of only one particular amino acid residue of a specific residue type at a specific sequence position, and

each non-linear term is a cross-product term comprising the product of a coefficient and two or more bit-value independent variables, wherein the coefficient of the non-linear term indicates the relative impact on activity by the interaction of the two or more bit-value independent variables, and wherein each of the two or more bit-value independent variables specifies the presence or absence of a particular residue of a specific residue type at a specific sequence position; and

(c) using the sequence-activity model to identify one or more amino acid residues at specific positions for varying or fixing to impact the desired activity.

51. The system of claim 50, further comprising a robotic liquid control armature for high-throughput liquid transferring of solutions from a source to a destination, wherein the robotic liquid control armature is operably linked to the control logic.

52. The system of claim 50, further comprising an oligonucleotide synthesizer or an oligonucleotide synthesis controller configured to synthesize or control the synthesis of one or more oligonucleotides having the one or more amino acid residues at the specific positions identified in (c), wherein oligonucleotide synthesizer or the oligonucleotide synthesis controller is operably linked to the control logic.

53. The system of claim 50, further comprising a massively parallel screening apparatus for assaying protein variants.

54. The system of 50, wherein the massively parallel screening apparatus comprises biosensors for detecting reaction product(s) selected from the group consisting of antibodies with reporter properties, and those based on in vivo affinity recognition coupled with expression and activity of a reporter gene.

55. The system of 50, wherein receiving data characterizing the training set of the protein variant library comprises receiving data from a client computer through a computer network.