Systems and methods for antibody engineering

Abstract

Methods, computer systems, and computer program products for biopolymer engineering. A variant set for a biopolymer of interest is constructed by identifying, using a plurality of rules, a plurality of positions in the biopolymer of interest and, for each respective position in the plurality of positions, substitutions for the respective position. The plurality of positions, and the substitutions for each respective position in the plurality of positions collectively defined a biopolymer sequence space. A variant set comprising a plurality of variants of the biopolymer of interest is selected. A property of all or a portion of the variants in the variant set is measured. A sequence-actively relationship is modeled between (i) one or more substitutions at one or more positions of the biopolymer of interest represented by the variant set and (ii) the property measured for all or the portion of the variants in the variant set. The variant set is redefined to comprise variants that include substitutions in the plurality of positions that are selected based on function of the sequence-activity relationship.

1 . A method for constructing a variant set for an antibody of interest, the method comprising:
 a) identifying, using a plurality of rules, a plurality of positions in said antibody of interest and, for each respective position in said plurality of positions, one or more substitutions for the respective position, wherein the plurality of positions and the one or more substitutions for each respective position in the plurality of positions collectively define an antibody sequence space;   b) selecting a variant set, wherein said variant set comprises a plurality of variants of said antibody of interest and wherein said variant set is a subset of said antibody sequence space;   c) measuring a property of all or a portion of the variants in said variant set;   d) modeling a sequence-activity relationship between (i) one or more substitutions at one or more positions of the antibody of interest represented by the variant set and (ii) the property measured for all or said portion of the variants in the variant set; and   e) redefining said variant set to comprise variants that include substitutions in said plurality of positions that are selected based on a function of said sequence-activity relationship.   
     
     
         2 . The method of  claim 1 , the method further comprising repeating said measuring, modeling, and, optionally, said redefining, until a variant in said variant set exhibits a value for said property that exceeds a predetermined value. 
     
     
         3 . The method of  claim 2  wherein said predetermined value is a value that is greater than the value for the property that is exhibited by said antibody of interest 
     
     
         4 . The method of  claim 1 , the method further comprising repeating said measuring, modeling, and, optionally, said redefining, until a variant in said variant set exhibits a value for said property that is less than a predetermined value. 
     
     
         5 . The method of  claim 4  wherein said predetermined value is a value that is less than the value for the property that is exhibited by said antibody of interest. 
     
     
         6 . The method of  claim 1 , the method further comprising repeating said measuring, modeling, and, optionally, said redefining, a predetermined number of times. 
     
     
         7 . The method of  claim 6  wherein said predetermined number of times is two, three, four, or five 
     
     
         8 . The method of  claim 1  wherein said sequence-activity, relationship comprises a plurality of values and wherein each value in said plurality of values describes a relationship between (i) a substitution at a position in said plurality of positions represented by said all or said portion of the variants in said variant set and said property, (ii) a plurality of substitutions at a position in said plurality of positions represented by said all or said portion of the variants in said variant set and said property, or (iii) one or more substitutions in one or more positions in said plurality of positions represented by said all or said portion of the variants in said variant set and said property. 
     
     
         9 . The method of  claim 8  wherein said modeling comprises regressing:
     V   measured   =W   11   P   1   S   1   +W   12   P   1   S   2   ++W   1N   P   1   S   N   ++W   M1   P   M   S   1   +W   M2   P   M   S   2 +W MN   P   M   S   N    
 wherein,
 V measured  represents the property measured in variants in said variant set; 
 W MN =is a value in said plurality of values; 
 P M =is a position in said antibody of interest in said plurality of positions in said antibody of interest; and 
 S N =is a substitution in the one or more positions for a position in the plurality of positions in said antibody of interest. 
 
 
     
     
         10 . The method of  claim 9  wherein said regressing comprises linear regression, nonlinear regression, logistic regression, multivariate data analysis, or partial least squares projection to latent variables. 
     
     
         11 . The method of  claim 1  wherein said modeling comprises computation of a neural network, computation of a bayesian model, a generalized additive model, a support vector machine, or classification using a regression tree. 
     
     
         12 . The method of  claim 1  wherein said modeling comprises boosting or adaptive boosting. 
     
     
         13 . The method of  claim 1  wherein said redefining further comprises:
 computing a predicted score for a population of variants of said antibody of interest using said sequence-activity relationship, wherein each variant in said population of variants includes a substitution at one or more positions in said plurality of positions in said antibody of interest; and 
 selecting said variant set from among said population of variants as a function of the predicted score received by each variant in said set of variants. 
 
     
     
         14 . The method of  claim 13 , the method further comprising
 ranking said population of variants, wherein each variant in said population of variants is ranked based on the predicted score received by the variant based upon the sequence-activity relationship; and   said selecting comprising accepting a predetermined percentage of the top ranked variants in said population of variants for said variant set.   
     
     
         15 . The method of  claim 13 , wherein a respective variant in said population of variants is selected for said variant set when the predicted score of the respective variant exceeds a predetermined value. 
     
     
         16 . The method of  claim 1  wherein said redefining step (e) further comprises redefining said variant set to comprise one or more variants each having a substitution in a position in said plurality of positions not present in any variant in the variant set selected by said selecting step (b). 
     
     
         17 . The method of  claim 1  wherein
 said modeling a sequence-activity relationship (d) further comprises modeling a plurality of sequence-activity relationships, wherein each respective sequence-activity relationship in said plurality of sequence-activity relationships describes the relationship between (i) one or more substitutions at one or more positions of the antibody of interest represented by the variant set and (ii) the property measured for all or said portion of the variants in the variant set; and 
 said redefining said variant set (e) comprises redefining said variant set to comprise variants that include substitutions in said plurality of positions that are selected based on a combination of said plurality of sequence-activity relationships. 
 
     
     
         18 . The method of  claim 17 , the method further comprising:
 repeating said measuring based upon said redefined variant set, wherein a property of all or a portion of the variants in the redefined variant set is measured; and weighting each respective sequence-activity relationship in said plurality of sequence activity relationships based on an agreement between (i) measured values for the property of variants in said redefined variant set and (ii) values for the property of variants in said redefined variant set that were predicted by said respective sequence-activity relationship, wherein   a first sequence-activity relationship that achieves better agreement between measured and predicted values than a second sequence-activity relationship receives a higher weight than said second sequence-activity relationship.   
     
     
         19 . The method of  claim 17  wherein said redefining step (e) further comprises redefining said variant set to comprise one or more variants each having a substitution in a position in said plurality of positions not present in any variant in the variant set selected by said selecting step (b). 
     
     
         20 . The method of  claim 18  wherein said redefining step (e) further comprises redefining said variant set to comprise one or more variants each having a substitution in a position in said plurality of positions not present in any variant in the variant set selected by said selecting step (b).