Methods, systems and devices for predicting PK value of antibodies

Abstract

Methods, systems, and devices are described for generating a pharmacokinetic (PK) model for predicting a pharmacokinetic (PK) value of an antibody of interest. An input dataset of a plurality of antibodies is received. The input data set includes data representing an amino acid sequence of each of the antibodies and an experimentally determined PK value of each of the antibodies. One or more surface properties of each of the antibodies are calculated based on these corresponding amino acid sequences. One or more region surface properties of one or more regions of interest are calculated for each of the antibodies based on the one or more surface properties calculated for each of the antibodies. A packet is determined from the one or more regions of interest that results in a maximum correlation between the corresponding calculated region surface properties of the packet and the corresponding PK value determined by the experiment. The method is used for establishing the PK region surface property relationship of various antibodies. A PK model for predicting the PK value of the antibody of interest is generated based on the PK region surface property relationship. The PK model is configured to receive one or more input region surface properties of the determined grouping of the antibodies of interest, and to output a predicted PK value for the antibodies of interest by applying the input region surface properties to the relationship.

Description

[0001] The present specification relates to methods, systems and devices for predicting the pharmacokinetic (PK) values ​​of antibodies prior to in vitro or in vivo analysis. Background Art

[0002] The introduction of therapeutic monoclonal antibodies (mAbs) into clinical practice has revolutionized healthcare, becoming the best-selling form of medicine in the past few years. The increasing popularity of antibody-based therapies has led to the development of next-generation multispecific antibody therapeutics that integrate two or more variable regions into a single molecule. Compared to combinations of monospecific antibodies, multispecific antibodies can bind two or more antigens simultaneously due to their architecture, which provides new opportunities to address complex diseases. To date, most of the available information is derived from monospecifics.

[0003] Recently, the pharmaceutical industry has shown great interest in revealing the pharmacokinetic (PK) principles of antibodies, because these principles provide key information about drug efficacy and dosing strategies. During antibody discovery / optimization activities, candidates with unacceptable PK properties are abandoned (or redesigned). Due to budgetary constraints and ethical issues, in vivo testing cannot be performed for a large number of candidates. Therefore, de-risking before performing in vivo PK assessments is highly desirable.

[0004] Therefore, multiple efforts have been made to associate in vivo PK data with in vitro properties and computer descriptors. The literature describes a series of experimental determinations of mAb and estimated biophysical properties that can be used to predict the PK of monospecific antibodies. These include, for example, multispecific reagent binding analysis, affinity capture self-interaction nanoparticle spectrum, the calculation of the surface properties of the combination with neonatal Fc receptors (FcRn) and antibodies. It is reported that mAbs that are excessively positive on the surface are more frequently internalized into cells by nonspecific pinocytosis, and show an increase in binding affinity with FcRn, thereby preventing the dissociation of the complex. Therefore, positive charge may interfere with the IgG recycling path and have a negative impact on PK. However, since this hypothesis is derived from conventional mAbs, its significance to multispecific forms is still unclear.

[0005] Accordingly, it is desirable to have an efficient, accurate and rapid mechanism or device to predict the PK value of any antibody of interest, thereby enabling preliminary screening of multiple antibodies that have not been tested or have unknown PK values. In particular, it is desirable to have an efficient, accurate and rapid mechanism or device to predict the PK value of a multispecific antibody of interest, thereby enabling preliminary screening of multiple multispecific antibodies that have not been tested or have unknown PK values. Summary of the invention

[0006] According to a first aspect of the present specification, there is provided a computer-implemented method for generating a PK model for predicting the pharmacokinetic (PK) value of an antibody of interest, the method comprising: receiving an input data set of multiple antibodies, the input data set comprising data representing an amino acid sequence of each of the antibodies and an experimentally determined PK value of each of the antibodies; calculating one or more surface properties of each of the antibodies based on the corresponding amino acid sequences; calculating one or more region surface properties of one or more regions of interest of each of the antibodies based on the one or more surface properties calculated for each of the antibodies; determining from the one or more regions of interest a grouping that produces a maximum correlation between the corresponding calculated region surface properties of the grouping and the corresponding experimentally determined PK value to establish a PK region surface property relationship for the multiple antibodies; and generating a PK model for predicting the PK value of the antibody of interest, wherein the PK model is configured to receive one or more input region surface properties of the determined grouping of the antibodies of interest, and output a predicted PK value of the antibody of interest by applying the input region surface properties to the PK region surface property relationship.

[0007] The computer-implemented method may further include a PK model, which is further configured to: calculate one or more surface properties of the antibody of interest based on the amino acid sequence of the antibody of interest; and calculate one or more region surface properties of the determined region of interest group of the antibody of interest based on the one or more surface properties calculated for the antibody of interest, thereby providing the input region surface properties.

[0008] The computer-implemented method, wherein the generated PK model is further configured to predict a shortlist of candidate antibodies having desired PK properties for use in in vitro wet lab analysis or in in vivo testing.

[0009] The computer-implemented method, wherein, for each of these candidate antibodies, the method further comprises the following steps: calculating one or more surface properties of each of the candidate antibodies based on the corresponding amino acid sequence of each of the candidate antibodies; and calculating one or more regional surface properties from the determined groupings of regions of interest for each of the candidate antibodies; inputting data representing the calculated regional surface properties of each of the candidate antibodies into the generated PK model for predicting the PK value of each of the candidate antibodies; receiving the predicted PK value of each of the candidate antibodies as output from the PK model; and when the received predicted PK value of the candidate antibody indicates one or more of the desired PK properties defined for the candidate antibody shortlist, adding the candidate antibody to the shortlist; and outputting data representing the candidate antibody shortlist for use at least in an in vivo test or in vitro wet lab analysis.

[0010] The computer-implemented method, wherein calculating the one or more surface properties of each of the plurality of antibodies comprises modeling the three-dimensional molecular structure of each of the antibodies and calculating the distribution of the one or more surface properties on the surface of the modeled molecular structure of each of the antibodies.

[0011] The computer-implemented method, wherein the calculated one or more surface properties include one or more of a positively charged surface area, a negatively charged surface area, and a hydrophobic surface area.

[0012] The computer-implemented method, wherein the surface of the modeled molecular structure of each of the antibodies comprises a plurality of patches having one or more surface properties, each patch having a patch area based on the distribution of the surface properties in the modeled molecular structure, wherein the number of patches for each antibody is the same as or different from the number of patches for each other antibody in the plurality of antibodies.

[0013] The computer-implemented method, wherein calculating the one or more region surface properties of the one or more regions of interest for each of the antibodies is based on an area of ​​the patch having the one or more surface properties associated with each region of interest.

[0014] The computer implemented method, wherein the plurality of antibodies and the antibody of interest are crossed dual variable (CODV) antibodies.

[0015] The computer-implemented method, wherein at least one region of interest is selected from the complementary domain region CDR, framework region or linker of the variable heavy VH domain or the variable light VL domain, including CDR1, CDR2, CDR3, FW1, FW2, FW3, FW4 of any one of the VH domain or VL domain of the two variable V domains of these CODV antibodies.

[0016] The computer implemented method, wherein the groupings from the regions of interest are CDR1 of VL1, CDR3 of VL1, and FW1 to FW4 of VL2 and VH2 of the CODV antibodies.

[0017] The computer implemented method, wherein the PK surface property relationship is a linear PK surface property relationship.

[0018] According to a second aspect of the present specification, a device is provided, comprising a processor, a memory unit and a communication interface, wherein the processor is connected to the memory unit and the communication interface, wherein the processor and the memory are configured to implement a computer-implemented method according to any one of the features or steps of the first aspect.

[0019] According to a third aspect of this specification, there is provided a computer-readable medium comprising data or instruction codes which, when executed on a processor, cause the processor to implement a computer-implemented method of any one of the features or steps of the first aspect.

[0020] According to a fourth aspect, a system is provided, comprising: a three-dimensional surface property module configured to receive data representing a plurality of candidate antibodies and calculate the regional surface properties of the region of interest of each of the candidate antibodies; a pharmacokinetic (PK) model module configured to receive the calculated regional surface properties corresponding to each of the candidate antibodies for predicting the PK value of each of the candidate antibodies; a PK comparison module configured to compare the predicted PK values ​​of the candidate antibodies with the expected PK properties and select those of the plurality of candidate antibodies having PK values ​​that satisfy the one or more expected PK properties; and an output module configured to output a shortlist of candidate antibodies from the selected candidate antibodies based on the comparison for use in in vitro wet lab analysis and / or in vivo testing.

[0021] The system may further be configured to implement one or more of the method steps or features according to the first aspect.

[0022] According to a fifth aspect of the present specification, a non-transitory tangible computer-readable medium is disclosed, comprising data or instruction code for generating a PK model for predicting the PK value of an antibody of interest, which, when executed on one or more processors, causes at least one of the processors to perform at least one of the steps of the method: receiving an input data set of multiple antibodies, the input data set comprising data representing an amino acid sequence of each of the antibodies and an experimentally determined PK value of each of the antibodies; calculating one or more surface properties of each of the antibodies based on the corresponding amino acid sequences; calculating one or more region surface properties of one or more regions of interest of each of the antibodies based on the one or more surface properties calculated for each of the antibodies; determining from the one or more regions of interest a grouping that produces the maximum correlation between the corresponding calculated region surface properties of the grouping and the corresponding PK value determined experimentally to establish a PK region surface property relationship of the multiple antibodies; and generating a PK model for predicting the PK value of the antibody of interest, wherein the PK model is configured to receive one or more input region surface properties of the determined grouping of the antibodies of interest, and output a predicted PK value of the antibody of interest by applying the input region surface properties to the PK region surface property relationship.

[0023] The non-transitory tangible computer readable medium may be further configured to implement one or more of the method steps or features according to the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to make the present invention more easily understood, embodiments of the present invention will now be described by way of example only with reference to the accompanying drawings, in which:

[0025] Figure 1a An example PK model pipeline process according to some embodiments of the present invention is shown;

[0026] Figure 1b An example PK model process according to some embodiments of the present invention is presented;

[0027] Figure 1c An example antibody preliminary screening process according to some embodiments of the present invention is presented;

[0028] Figure 2a An example CODV antibody structure and an example three-dimensional (3D model) of a CODV antibody according to some embodiments of the present invention are shown;

[0029] Figure 2b Further exemplary CODV antibody structures according to some embodiments of the present invention are shown;

[0030] Figure 2cpresents an example multi-specific CODV Ig-like dataset according to some embodiments of the present invention;

[0031] Figure 3a A system 300 for generating and using a PK model according to some embodiments of the present invention is shown;

[0032] Figure 3b A computer computation pipeline according to some embodiments of the present invention is presented;

[0033] Figure 3c The regions of interest of the CODV antibody structure according to some embodiments of the present invention are shown;

[0034] Figure 4a A set of plots showing correlations between experimental PK clearance values ​​and ion properties in two regions of interest according to some embodiments of the present invention;

[0035] Figure 4b A set of plots showing the correlation between experimental PK clearance values ​​and hydrophobic properties in a region of interest according to some embodiments of the present invention;

[0036] Figure 4c Some embodiments of the present invention are shown Figure 4a and Figure 4b Cluster drawing of relationships between plots;

[0037] Figure 4d Some embodiments of the present invention are shown Figure 4a and Figure 4b Correlation plot of the combined scores (specified as in silico clearance likelihood) of the plot;

[0038] Figure 5 Plots associated with two tested CODV antibodies, both of which include the same V domains but with their orientations swapped, plotting the ratio of cationic ions in CDR1 and CDR3 of VL1 versus the hydrophobic area in FW of VR2;

[0039] Figure 6 Plots associated with two tested CODV antibodies, both comprising the same V domains but with their orientations swapped, plotting the ratio of cationic ions in CDR1 and CDR3 of VL1 versus the hydrophobic area in FW of VR2 (left) and plotting the combined score (assigned as in silico clearance likelihood) versus experimentally determined in vivo clearance (right);

[0040] Figure 7shows a correlation plot of the combined scores (designated as in silico clearance likelihood) of a validation set of seven new bispecific CODV molecules, whose PK properties were first predicted using the PK model pipeline of the present invention and whose in vivo PK (clearance) was later assessed;

[0041] Figure 8 is a schematic diagram of a system / apparatus for performing the methods described herein.

[0042] Common reference numerals are used throughout the drawings to indicate similar features. DETAILED DESCRIPTION

[0043] Various example embodiments described herein relate to (multiple) methods, devices and (multiple) systems for automatically, efficiently and reliably performing computer prediction or estimation of the PK value of an antibody using a PK model generated from experimental pharmacokinetic (PK) data of a group of antibodies. PK values ​​of interest include, but are not limited to, for example, PK clearance or clearance (unit mL / h / kg), PK half-life (T1 / 2) and PK distribution volume (Vss) and / or any other suitable PK value of interest. The PK model is generated from a computational pipeline that estimates the surface patch landscape of an antibody (e.g., a crossover dual variable (CODV) antibody and / or any other type of antibody), analyzes the landscape with experimental PK data available for a group of known antibodies, and identifies a group of regions of interest that are most relevant to the experimental PK data on the landscape to form a PK value relationship.

[0044] The PK model is formed by the identified grouping of the PK value relationship and the region of interest. Once formed, the PK model can be used to identify the candidate antibodies of interest to generate a final candidate list of antibodies that meet certain PK value requirements (e.g., fast or slow in vivo clearance), and then the antibodies screened initially are subjected to in vitro wet laboratory analysis and / or in vivo testing. The surface patch landscape includes a plurality of surface patches (e.g., regions with specific surface properties on the surface of the antibody), each surface patch having a surface patch property, including but not limited to at least one of, for example, ionic properties and hydrophobic properties.

[0045] Depending on the application requirements, the regions of interest of an antibody may include, but are not limited to, for example, an antigen binding fragment (Fab) region, one or more portions of the variable binding region of a Fab; multiple portions of the variable binding region associated with variable domain 1 (VR1) or 2 (VR2); multiple complementarity determining regions (CDRs) corresponding to each of the VR1 domain and VR2 domain of an antibody; multiple framework (FW) regions corresponding to each of the VR1 domain and VR2 domain of an antibody; one or more portions of the variable heavy chain (VH) region and / or the variable light chain (VL) binding region of the VR1 and / or VR2 binding regions of an antibody; one or more portions of the VH1, VH2, VL1, and VL2 binding regions of the VR1 and / or VR2 binding regions of an antibody; multiple CDR regions and multiple FW regions associated with each of the VH1, VH2, VL1, and VL2 binding regions of an antibody; one or more linkers (L1 to L4) of an antibody molecule; and / or any other combination of regions of interest of an antibody.

[0046] Embodiments of PK model as described herein and / or process provide PK value (for example, PK clearance rate) efficient and accurate computer prediction and estimation rather than perform expensive and time-consuming external and / or in vivo test advantage.A further advantage of PK model as described herein is that the quantity of possible candidate antibodies and / or CODV antibodies with unknown PK is reduced and efficiently preliminary screening candidate antibodies or CODV antibodies makes only the most promising candidate antibodies or candidate CODV antibodies be selected for external wet laboratory analysis or the ability of screening and / or in vivo test.Therefore, before expensive and / or time-consuming external wet laboratory analysis and / or subsequent in vivo test, only these antibodies of estimation / prediction PK value / characteristic (for example, PK clearance rate) that meet expectation or required PK threshold value or characteristic can be preliminarily screened.

[0047] The method of the present application allows for the first time robust PK predictions for multispecific (e.g., bispecific) antibodies. In addition, it successfully associates the size of plaques with surface properties with PK values. About CODV antibodies, it is determined that the orientation of V1 and V2 has an impact on PK values.

[0048] Figure 1a An example PK model pipeline process 100 for generating a PK model for predicting PK values ​​of an antibody of interest is shown. The PK model pipeline process 100 includes at least the following steps:

[0049] In step 101, an input dataset of a plurality of antibodies is received, the input dataset comprising data representing an amino acid sequence of each of the plurality of antibodies and an experimentally determined PK value (eg, PK clearance rate) of each of the antibodies.

[0050] In some embodiments of the invention, the plurality of antibodies and the antibody of interest are multispecific antibodies.

[0051] In some embodiments of the invention, the various antibodies and antibodies of interest are CODV antibodies. In some embodiments of the invention, the various antibodies and antibodies of interest are tetravalent bispecific CODV antibodies. In some embodiments of the invention, the various antibodies and antibodies of interest are trivalent trispecific CODV antibodies. In some embodiments of the invention, the various antibodies and antibodies of interest are bivalent bispecific CODV antibodies.

[0052] In step 102, one or more surface properties of each of the plurality of antibodies are calculated based on the corresponding amino acid sequence of each of the plurality of antibodies.

[0053] For example, calculating one or more surface properties of each of a plurality of antibodies may include, but is not limited to, for example, modeling the three-dimensional (3D) molecular structure of each of the antibodies and calculating the distribution of the one or more surface properties on the surface of the 3D modeled molecular structure of each of the antibodies.

[0054] In essence, several types of surface properties can be calculated for each of the antibodies, and these surface properties include, but are not limited to, for example, ionic surface properties or hydrophobic surface properties. Ionic surface properties include negatively charged and positively charged surface areas. Hydrophobic surface properties include hydrophobic surface areas. Therefore, the one or more surface properties of each antibody calculated include surface property areas based on one or more of positively charged surface areas, negatively charged surface areas, and hydrophobic surface areas on the surface of the 3D molecular structure of each antibody. The modeling of the 3D structure and the distribution of the one or more surface properties on the surface of the 3D structure can be repeated several times for each of the antibodies. Later, these repeated results can be averaged. In one embodiment, the modeling of the 3D structure and the distribution of the one or more surface properties on the surface of the 3D structure are repeated N times (e.g., N=50, N=100 or any other suitable sample size) and averaged later.

[0055] The surface of the modeled molecular structure of each of the antibodies includes one or more surface property regions having one or more surface properties, each surface property region including a plurality of patches of the one or more surface properties. Each patch has a patch area based on the distribution of the surface properties in the modeled molecular structure, wherein the number of patches of each antibody is the same as or different from the number of patches of each other antibody in the plurality of antibodies. Each patch is a negatively charged patch; a positively charged patch; or a hydrophobic patch.

[0056] In another example, the at least one surface property of the plurality of patches on the surface of each antibody comprises an ionic surface property, and the ionic surface property of each patch comprises a positively charged ionic region or a negatively charged ionic region.

[0057] In step 103 , one or more region surface properties of one or more regions of interest for each of the antibodies are calculated based on the one or more surface properties calculated in step 102 for each of the antibodies.

[0058] For example, each region of interest on the surface of the modeled molecular structure of each of the antibodies comprises one or more surface regions with one or more surface region properties, and each surface property region comprises one or more patches or multiple patches. The one or more region surface properties of the one or more regions of interest of each of the antibodies are calculated to be the area of ​​the patches based on the one or more surface properties. Each region of interest can include at least three types of surface properties, including but not limited to, for example, negatively charged regions, positively charged regions, and hydrophobic regions.

[0059] In another example, calculating at least one region surface property corresponding to one or more regions of interest for each of the antibodies may further include calculating, for each of the antibodies, at least one region surface property for each region of interest based on identifying those regions associated with each region of interest for each antibody and combining the corresponding surface properties of the identified regions for each antibody.

[0060] In another example, the at least one region surface property corresponding to one or more regions of interest includes an ion surface property corresponding to one or more regions of interest. The ion surface property of each region of interest may include a positive ion region surface property and a negative ion region surface property, wherein calculating the region surface property of each of these regions of interest for each of the antibodies further includes calculating the ion region surface property of each of these regions of interest for each of the antibodies based on: calculating the positive ion region surface property of each region of interest for each of the antibodies by aggregating or averaging the patches identified as being associated with each region of interest having a positively charged ion region; and calculating the negative ion region surface property of each region of interest for each of the antibodies by aggregating or averaging the patches identified as being associated with each region of interest having a negatively charged ion region.

[0061] In step 104, a grouping is determined from the one or more regions of interest that yields the greatest correlation between the corresponding calculated region surface properties of the grouping and the corresponding experimentally determined PK values. The correlations can be analyzed to establish a PK region surface property relationship for the plurality of antibodies.

[0062] For example, a linear regression can be performed for each grouping from one or more regions of interest for all of the multiple antibodies based on: plotting data points that correspond to the corresponding calculated region surface properties of the groupings of each antibody and the corresponding PK values ​​determined experimentally for the each antibody; and performing a linear regression analysis on the resulting data points. The grouping that produces the greatest correlation (e.g., the greatest positive or negative correlation) based on the linear regression is then selected as the determined grouping. The linear regression output of the selected grouping is used to establish the PK region surface property relationship for the multiple antibodies.

[0063] Determining the grouping from the one or more regions of interest can further include determining the grouping that produces the maximum correlation (e.g., positive or negative correlation) between the combined region surface properties of the regions of interest of the antibodies and the corresponding experimentally determined PK values ​​of the antibodies based on the combination of the one or more regions of interest of the antibodies. Establishing the PK surface property relationship can be based on estimating the correlation relationship between the determined groupings of the antibodies and the corresponding experimentally determined PK values. For example, establishing the PK surface property relationship can include calculating a linear PK surface region property relationship based on associating the combined region surface properties of the determined groupings of each of the multiple antibodies with the corresponding experimentally determined PK values ​​of each of the multiple antibodies.

[0064] Alternatively or additionally, nonlinear regression can be performed for each grouping from one or more regions of interest for all of the multiple antibodies based on: plotting data points that correspond the corresponding calculated region surface properties of the groupings of each antibody to the corresponding PK values ​​determined experimentally for the each antibody; and performing nonlinear regression analysis on the resulting data points. The grouping that produces the greatest correlation based on linear regression is then selected as the determined grouping. The linear regression output of the selected grouping is used to establish the PK region surface property relationship for the multiple antibodies. In any case, a PK region surface property relationship is established.

[0065] In step 105, a PK model for predicting a PK value of an antibody of interest is generated based on the determined grouped PK region surface property relationships associated with the maximum correlation. The PK model is configured to predict a PK value of the antibody of interest by receiving one or more input region surface properties of the determined grouped antibodies of interest, processing the determined grouped input region surface properties of the antibody of interest according to the PK region surface property relationships, and outputting a predicted PK value of the antibody of interest. The processing may include applying the input region surface properties to the PK region surface property relationships.

[0066] The PK model can be further configured to receive data representing an antibody of interest, such as its amino acid sequence, and calculate one or more surface properties of the antibody of interest based on the amino acid sequence of the antibody of interest. The PK model can then calculate one or more region surface properties of the determined groupings of regions of interest from the antibody of interest based on the one or more surface properties calculated for the antibody of interest, thereby providing the input region surface properties.

[0067] The PK model may also be configured to assign different PK markers to two or more non-overlapping PK value ranges. The PK model may further be configured to determine which of the two or more PK ranges the predicted PK value falls into and output a corresponding PK marker and / or predicted PK value. For example, when the PK value is a PK clearance rate, the marker may correspond to, but is not limited to, for example, a "slow," "moderate," or "fast" PK clearance rate, or any other PK clearance rate, etc.

[0068] As an example, the generated PK model is further configured and used to predict a shortlist of candidate antibodies with desired PK properties for use in in vitro wet laboratory analysis and / or in vivo testing. Each candidate antibody is input into the PK model as an antibody of interest, for example, the amino acid sequence of the candidate antibody is input, wherein the group region surface properties are calculated based on the amino acid sequence of the candidate antibody and applied to the PK region surface property relationship. When the estimated PK value is higher or lower than a certain PK threshold value (depending on what is desired) or the PK value associated with the group region surface property is aggregated in the PK value region associated with the preliminary screened candidate antibody, the candidate antibody exhibits the desired PK properties.

[0069] Figure 1b An example PK model process 110 for predicting PK values ​​for an antibody of interest is shown. Assume that the PK model has been developed based on Figure 1a The PK model pipeline process 100 is generated. The PK model process 110 includes the following steps:

[0070] In step 111, data representing an antibody of interest, such as its amino acid sequence, is received.

[0071] In step 112, one or more surface properties of the antibody of interest are calculated based on the amino acid sequence of the antibody of interest. Calculating the one or more surface properties of the antibody of interest based on the amino acid sequence of the antibody of interest may further include modeling the 3D molecular structure of the antibody of interest based on the amino acid sequence of the antibody of interest and calculating the distribution of the one or more surface properties on the surface of the modeled molecular structure of the antibody of interest.

[0072] In step 113, one or more region surface properties of the antibody of interest are calculated based on the calculated one or more surface properties of the antibody of interest using the region groupings of interest determined from step 104. The region surface properties are thereby provided as input for application to a PK-region surface property relationship for estimating a PK value.

[0073] In step 114, the PK values ​​of the antibodies of interest are estimated based on the calculated zone surface properties using the PK model generated in step 105 by inputting data representing the calculated zone surface properties into the generated PK model for predicting the PK value of each of the antibodies of interest.

[0074] In step 115, an indication of the PK value of the antibody of interest is output based on the estimated PK value.

[0075] Figure 1c An example antibody preliminary screening process 120 is shown, using the PK model process 110 to predict a shortlist of antibodies that have PK values ​​in the desired range or meet the desired PK characteristics / properties. Assume that the PK model has been Figure 1a The PK model pipeline process 100 is generated. The antibody preliminary screening process 120 includes the following steps:

[0076] In step 121, data representing multiple antibodies with unknown PK values ​​are received. The data representing each antibody may include, but is not limited to, for example, the amino acid sequence of the antibody or other standard molecular structures representing the antibody.

[0077] In step 122, the surface properties of each antibody are calculated based on the amino acid sequence of the antibody or other standard molecular structures representing the antibody. For example, the surface properties of each antibody can be calculated based on the following: N 3D simulations are performed on each antibody to calculate the surface properties of each antibody based on the amino acid sequence of the antibody or other standard molecular structures representing the antibody. A set of surface properties of the antibody is generated for each 3D simulation of the antibody, and the N sets of surface properties generated from the 3D simulations of each antibody are aggregated and averaged.

[0078] In step 123, each antibody in the plurality of antibodies is processed to determine whether the antibody meets the PK criteria / properties for inclusion in the shortlist. Each candidate antibody from the plurality of antibodies is selected for testing / evaluation based on the following steps for each antibody:

[0079] In step 124, the calculated regional surface properties associated with the selected candidate antibody as the antibody of interest are applied to a PK model configured to predict or estimate the PK value of the antibody (e.g., Figure 1a The process of 100 or Figure 1bPK model 110 outputs the PK model).

[0080] In step 125, data output from the PK model representing a predicted or estimated PK value for the antibody of interest is received.

[0081] In step 126, it is determined whether the output received PK value of the antibody of interest is within the desired PK value range required for the preliminary screening of each antibody. If the received PK value is within the expected PK value range, or reaches above the expected minimum PK threshold or below the expected PK maximum threshold (e.g., Y), the antibody of interest meets the PK requirement and the process 120 proceeds to step 128. If the received PK value is outside the expected PK value range, or reaches below the expected minimum PK threshold or above the expected PK maximum threshold (e.g., Y), the antibody of interest does not meet the PK requirement and the process 120 proceeds to step 127.

[0082] In step 127, the next candidate antibody in the plurality of candidate antibodies is selected as the antibody of interest, and the process 120 proceeds to step 124 to test the selected antibody of interest.

[0083] In step 128, when the received predicted PK value of the candidate antibody indicates one or more of the desired PK properties / ranges / thresholds defined for the shortlist of candidate antibodies, the antibody of interest is added to the shortlist as the candidate antibody.

[0084] Once all candidate antibodies in the plurality of antibodies have been processed, data representing a shortlist of candidate antibodies is output for use in at least in vitro wet lab assays and / or in vivo trials to analyze efficacy and / or PK of antibodies of that type.

[0085] Figure 1a to Figure 1c Each of the processes 100, 110, 120 can be implemented on a device including a processor, a memory unit, and a communication interface. The processor is connected to the memory unit and the communication interface. The processor and the memory can be configured to implement the processes 100, 110, 120 and / or each of the other processes described herein as computer-implemented methods. The processes 100, 110, 120 and / or combinations thereof, modifications to these processes based on one or more other processes as described herein can be stored on a computer-readable medium. The computer-readable medium may include data or instruction codes that, when executed on a processor, cause the processor to implement the processes 100, 110, 120 and / or combinations thereof, modifications to these processes as one or more computer-implemented methods as described herein.

[0086] In certain embodiments, the multiple antibodies and the antibody of interest are cross double variable (CODV) antibodies. However, any type of antibody can be used. It may be preferred that the multiple antibodies used to generate the PK model belong to the same antibody family or the same type of antibody. It may be preferred that the antibody form of the multiple antibodies used to generate the PK model belongs to the same antibody form. For example, only CODV antibodies are used to generate the CODV PK model, and only monoclonal antibodies (mAbs) are used to generate the mAb PK model. For example, only multispecific antibodies are used to generate multispecific PK models and only monospecific antibodies are used to generate monospecific PK models.

[0087] As used herein, "CODV antibody" includes all antibodies or antibody fragments comprising two V domains with a cross orientation. Previously, CODV antibodies have been described in international patent applications WO2012 / 135345 and WO20161 / 16626 and the publications of Steinmetz et al. (MAbs.2016 July; 8(5): 867-78), which are incorporated herein by reference in their entirety. In certain embodiments, CODV antibody is in the form of CODV-Ig, i.e. including the Fc domain. In certain embodiments, CODV antibody is in the form of CODV-Fc-OL, i.e. including the Fc domain but only one CODV-Fab. In certain embodiments, CODV antibody is in the form of CODV, including a CODV-Fab and a conventional Fab, thereby producing a trispecific structure.

[0088] CODV antibodies are multispecific. As used herein, "multispecific" refers to antibodies that specifically bind to more than one target. In some embodiments, CODV antibodies are bivalent and bispecific. In some embodiments, CODV antibodies are trivalent and bispecific. In some embodiments, CODV antibodies are trivalent and trispecific. In some embodiments, CODV antibodies are tetravalent and bispecific. In some embodiments, CODV antibodies are tetravalent and trispecific. In some embodiments, CODV antibodies are tetravalent and tetraspecific.

[0089] In some embodiments, for CODV antibodies, the region of interest can be selected from CDR1, CDR2, CDR3, FW1, FW2, FW3, FW4 of any one of the VH domain or VL domain of the two V domains of the CODV antibody. It can be found that the grouping from the region of interest that produces the greatest correlation is CDR1 of VL1 of CODV antibody, CDR3 of VL1, and FW1 to FW4 of VL2 and VH2.

[0090] Although the following description describes a PK model generation pipeline and application related to the clearance rate (or PK clearance rate) of a crossover dual variable (CODV) Ig-like form of an antibody, this is only by way of example and the present invention is not limited thereto. Those skilled in the art should understand that the principles described herein for generating a PK model can be applied to any type of antibody according to application requirements.

[0091] Figure 2a and Figure 2b are example CODV antibody architectures / structures 200 , 210 , 220 , 230 , and 240 , according to some embodiments. Figure 2a CODV antibody architecture 200 is a schematic diagram of a CODV antibody showing two VR binding domains (VR1, VR2) connected via a linker (L). VR1 and VR2 are paired together in a crossover manner, wherein the VH1 of VR1 is bound to the VH2 of VR2 via L3 and the VL1 of VR1 is bound to the VL2 of VR2 via L1. The VL1 of VR1 is bound to the constant domain (CL) of the light chain via L2, and the VH2 of VR2 is bound to the first constant domain (CH1) of the heavy chain via L4. VH1 is paired with VL1 (VR1), and VH2 is paired with VL2 (VR2). The CODV antibody architecture 210 is a 3D molecular model of the CODV molecule highlighting the VR1 and VR2 regions, which is generated using a molecular operating environment (MOE) (Molecular Operating Environment (MOE) [Molecular Operating Environment], 2020.09 Chemical Computing Group [Chemical Computing Group] ULC, 1010 Sherbooke St. West, Suite #910, Montreal, QC, Canada, H3A 2R7, 2022 [1010 Sherbooke St. West, Suite #910, Montreal, QC, Canada]) or other 3D modeling software.

[0092] Figure 2b CODV antibody framework 220 is another example CODV framework, which includes two VR binding domains paired together in a cross-wise manner as fragment antigen binding (Fab) arms of immunoglobulin G (IgG). This allows the design of bispecific (bi-Ab) and trispecific (tri-Ab) modalities shown as CODV antibody frameworks 220, 230 and 240. Three modalities 220, 230 and 240 contain at least one CODV-Fab arm and are different from each other in the identification of the second arm. Regarding the bispecific modality, CODV-IgG has two identical CODV-Fab arms, while CODV-FcOL lacks the second arm and is asymmetric. The second arm of the CODV trispecific scaffold shown is a standard Fab.

[0093] In the basic structure of CODV-Fab arm, the linear sequence of heavy chain (HC) and light chain (LC) is encoded for their corresponding contribution in two VR domains (named as VR1 and VR2) of molecule. Each VR is directed to an epitope of interest, which allows multi-specificity. About its structure, CODV-Fab arm is formed by inner domain and outer domain, and this inner domain and this outer domain are connected by two joints (L1 and L3) separating VR, and this allows the proper folding of molecule. The constant region of heavy chain, VR2 and the joint (L4) connecting them form the inner domain, and the outer domain is composed of the constant region of light chain, VR1 and their joint (L2).

[0094] Although several modalities 210 to 240 are described herein, this is only by way of example and the present invention is not limited thereto. Those skilled in the art should understand that other types of antibody architectures / forms can be applied to the PK pipeline and the corresponding PK models generated thereby according to application requirements.

[0095] Figure 2a and Figure 2b Different types of modalities are presented that can be used to form: CODV-based data sets with known PK values ​​or PK properties for generating PK models; and / or CODV data sets with unknown PK values ​​or PK properties for evaluation and / or screening using PK models to determine PK properties, followed by in vitro and / or in vivo testing of the screened candidate CODV antibodies.

[0096] Figure 2c A multi-specific CODV Ig-like dataset 250 is presented, comprising a plurality of CODV antibody molecules, including 30 different CODV Ig-like antibodies from a diverse selection of target specificities and formats presented in CODV dataset table 251. In this example, the CODV dataset consists of 30 CODV antibodies. 23 of the CODV antibodies are bispecific (comprising two CODV formats: CODV-IgG and CODV-FcOL) and 7 CODV antibodies are trispecific (combining a CODV-Fab and a standard Fab arm). Figure 2b The three CODV modes are shown in Figure 2.

[0097] The CODV clearance table 251 is about 30 CODV molecules specified by CODVID 1 to 30 and corresponding known PK clearance values ​​(mL / h / kg) (i.e., PK values). The CODV clearance distribution chart 252 shows the distribution of CODV clearance values ​​of CODV antibodies from the CODV clearance table 251. In this example, the PK clearance threshold 253 is set to 1 mL / h / kg shown by the dotted line, which can be used in this example to define whether the CODV antibody is in the slow clearance group or the fast clearance group of CODV 254 and 255, respectively.

[0098] That is, if the PK clearance value of CODV antibody molecule is higher than the PK clearance threshold, CODV antibody molecule is regarded as a part of fast clearance CODV antibody molecule or fast clearance group 255. Similarly, if the PK clearance value of CODV antibody molecule is at or below the PK clearance threshold, CODV antibody molecule is regarded as a part of slow clearance CODV antibody molecule or slow clearance group 254. It should be noted that this PK clearance threshold can be set to any other value and / or multiple PK clearance thresholds can be set to define grouping, multiple clearances, or PK clearances of different groups for marking / characterizing CODV antibody molecules. Other PK clearance thresholds can be set depending on the type of antibody, for example, in the literature for mAb, the PK clearance threshold in the region of 0.32mL / h / kg has been cited (Avery et al., MAbs.2018 February to March; 10 (2): 244–255), but this typically depends on the protocol used in determining the PK clearance value, and the method may vary with different thresholds. The PK clearance threshold may be set by the user of the system or defined by PK value requirements from in vitro wet lab analysis and / or in vivo testing that the candidate CODV molecule is required to meet.

[0099] Figure 2c The CODV dataset 250 is used herein as an example CODV dataset to demonstrate the pipeline process 100, the resulting PK model process 110, and the process 120 for selecting a candidate PK model with an unknown PK value associated with the PK model. Although an example CODV dataset of 20 CODV-like antibody molecules is described herein, this is only by way of example and the present invention is not limited thereto, and those skilled in the art will understand that any antibody dataset with multiple antibodies and known PK values ​​(e.g., clearance and / or other PK characteristics) can be used to generate a PK model as described herein and be associated for predicting or estimating corresponding PK values ​​of an unknown group of antibodies and / or one or more unknown antibodies, etc.

[0100] Figure 3aA system 300 for generating a PK model for predicting the PK value of an antibody is shown. The system 300 includes a 3D modeling module 301, a 3D surface property module 302, a PK model generation module 303, and a PK model module 304 connected together. The functions of the 3D modeling module 301, the 3D surface property module 302, the PK model generation module 303, and the PK model module 304 can be implemented Figure 1a , Figure 1b and / or Figure 1c The steps and / or functions of the PK model pipeline process 100, the PK model process 110 and the antibody preliminary screening process 120 are described in detail. In operation, the 3D modeling module 301 is configured to model the 3D structure of each antibody. For example, modeling the 3D structure of a group of CODV antibody molecules can be based on the amino acid sequence of the antibody molecules. The 3D surface property module 302 is configured to receive data representing the 3D structure of the antibody and calculate the surface properties of the region of interest of each antibody (e.g., steps 101, 102 and 103 of the PK model pipeline process 100). The PK model generation module 303 can be configured to generate a 3D model based on the 3D structure of each antibody. Figure 1a The PK model pipeline process 100 of the present invention is performed according to steps 104 and 105 to generate a PK model. The PK model module 304 includes one or more generated PK models, which are configured to receive the calculated regional surface properties of the antibody of interest from the 3D surface property module 302, and apply these regional surface properties to one of the PK models to predict or estimate the PK value of the antibody of interest. The system 300 may further include an output module for displaying or sending data representing the predicted or estimated PK value of the antibody of interest to an operator or user of the system 300.

[0101] PK model module 304 may further include a function (e.g., one or more steps of process 120) for selecting a candidate antibody final candidate list (e.g., CODV antibody) from a plurality of antibodies based on the desired PK characteristics / properties. In operation, each of the candidate antibodies is input to 3D modeling module 301 as an antibody of interest to model the 3D structure of every kind of antibody of interest. The 3D model of the antibody of interest is provided to 3D surface properties module 302 to calculate the region surface properties of the region of interest of every kind of antibody of interest. PK model module 304 receives the calculated region surface properties of the antibody of interest from 3D surface properties module 302, and applies these region surface properties to one of the PK models to predict or estimate the PK value of the antibody of interest. PK model module 304 may further include a PK comparison module (not shown), which is configured to compare the prediction or estimation PK value of the candidate antibody with the expected PK property and select those candidate antibodies with the PK value that meets the one or more expected PK properties in the plurality of candidate antibodies. The output module may be configured to output a shortlist of candidate antibodies from the selected candidate antibodies based on the comparison for use in in vitro wet lab analysis and / or in vivo testing.

[0102] Figure 3b A computer computational pipeline 310 is shown for use in system 300 to generate a PK model for predicting PK values ​​of an antibody of interest. The computational pipeline 310 is based on the PK model pipeline process 100, which has a Figure 1a The PK model pipeline process 100 is further modified by steps 101 to 105. In this example, the computational pipeline 310 is described with reference to CODV antibodies and basic CODV-Fab structures. The computational pipeline 310 is configured to generate a PK clearance model for predicting clearance in CODV antibodies. Although PK clearance and PK clearance models are described herein, this is only by way of example and the present invention is not limited thereto. Those skilled in the art will appreciate that the computational pipeline 310 and the resulting PK model can be applied to any type of antibody and PK, PK value or PK property according to application requirements. In this example, it is assumed that multiple CODV antibodies with known PK clearance values ​​are used to generate a PK clearance model. For example, as referenced Figure 2c The described CODV antibody molecule dataset 250 with CODVID 1 to 20 can be used as a plurality of CODV antibodies with known PK values ​​(e.g., PK clearance in this example). The computational pipeline 310 can include the following steps:

[0103] In step 311, the amino acid sequence of each in the CODV antibody from the CODV data set with known PK value is input to 3D modeling module 301, in which the 3D modeling system generates the 3D model of every kind of CODV antibody.For example, the FASTA format can be used to input the amino acid sequence.Although the FASTA format is described herein, this is only by way of example and the present invention is not limited thereto, and it will be appreciated by those skilled in the art that any other suitable amino acid sequence data format can be used according to application requirements.For example, according to application requirements, the software for generating each 3D model of every kind of CODV can be but is not limited to such as molecular operating environment (MOE) and / or any other suitable 3D modeling system or software.

[0104] For CODV-Fab structure, workflow starts with the homology model that generates CODV-Fab arm.Can be based on using MOE to perform homology modeling.Use chemical computing group (CCG) annotation numbering scheme to identify CDR, FW and joint from sequence.Although CCG annotation numbering scheme is described herein, this is only by way of example and the present invention is not limited thereto, and those skilled in the art will appreciate that any other suitable annotation numbering scheme format can be used according to application requirements.In five CODV-Fab arms (for example, PDB code: 6O8D, 5FHX, 5WHZ, 6O89 and 5HCG) of publicly available crystal structure, a CODV-Fab arm with the shortest joint length difference is used as template.After selecting the best CODV-Fab arm template, inner domain and outer domain are independently modeled and assembled with connecting joint.VR is modeled using the best template from MOE antibody database. Finally, the entire structure was protonated at a defined pH (eg, at physiological pH 7.4) and a set of minimization protocols (rigid body and free minimization) were applied using the AMBER10:EHT force field to produce a high-quality 3D model.

[0105] In step 312, the 3D modeling module 301 can be further configured (or the 3D modeling system can be configured) to model the 3D structure of each CODV antibody based on one or more regions of interest. In this case, the portions or regions of interest of the CODV antibodies being modeled are the CH1, VH2, VH1, CL, VL1, and VL2 domains and their connecting joints (L1, L2, L3, and L4). Figure 3c Other areas of interest are shown in .

[0106] In step 313, after generating the 3D model of each CODV antibody, the 3D model structure of each of the CODV antibodies in the data set is passed to the 3D surface property module 302, in which, for each of the CODV antibodies, the surface side chains (amino acids) of each 3D model of each CODV antibody are sampled (using MOE) to generate a set of structures of a patch surface map intended to accurately describe the 3D model structure of each CODV antibody. In this example, the number of samples N = 50 is used so that for each CODV antibody in a plurality of CODVs with known PK values ​​(e.g., Figure 2c CODVID 1 to 20) models 50 different sampling structures. Although this example describes the use of N = 50 samples, this is only by way of example and the present invention is not limited thereto. Those skilled in the art will appreciate that any suitable number N>0 samples may be used.

[0107] The plaque surface landscape may be affected by the conformation of the CODV side chains, which can adopt multiple energetically favorable orientations. Therefore, using a single structural model to analyze the properties of the molecular surface may not fully represent the reality. Therefore, a set of N=50 different conformers of each CODV-Fab arm model of each CODV antibody in data set 250 is generated by using a sampling protocol at a defined pH (e.g., at physiological pH 7.4). After generating the conformers, all surface plaques are calculated. The surface properties of the plaque may be hydrophobic, negatively charged or positively charged depending on its physicochemical properties.

[0108] Each patch has an associated surface area (in The CODV-Fab arm sequences were annotated with a list of residues (measured in square angstroms or square angstroms). All patches belonging to each FW, CDR and joint (16 FWs, 12 CDRs and 4 joints of a CODV-Fab arm) were identified using, but not limited to, the CCG annotation scheme or any other suitable annotation scheme. Next, the total hydrophobic region, negatively charged region and positively charged region from all FWs, CDRs and joints were obtained for each conformer. Finally, the average value of all conformer structures was calculated, and representative hydrophobic, ion-negatively charged and ion-positively charged patch areas (in terms of ) were obtained from each FW, CDR and joint on all calculated conformations of each CODV antibody in data set 250. This is performed in the following steps 314 to 315.

[0109] In step 314, the surface patch is calculated by the 3D surface property module 302 for each sampled structure of each CODV antibody, wherein the surface patch includes a positively charged patch; a negatively charged patch; or a hydrophobic patch. In other words, the surface patch includes a surface property including an ionic surface property or a hydrophobic surface property, and the ionic surface property includes an ionic positive charge and an ionic negative charge.

[0110] In step 315, the 3D surface property module 302 generates the surface patches (ion positive, ion negative and hydrophobic), their surface areas (in The surface properties of each of the regions of interest (e.g., framework (abbreviated as FW or FR), complementarity determining region (CDR), linker, or a combination of such regions) are determined by summing / adding the magnitudes of the surface properties of these patches that are co-located and / or overlapping with each region of interest.

[0111] As an example, in the case where the at least one surface property of the plurality of patches on the surface of each antibody is an ionic surface property, the ionic surface property of each patch includes a positively charged ionic region or a negatively charged ionic region. The ionic surface property of each region of interest includes a positive ion region surface property and a negative ion region surface property. Calculating the region surface property of each of these regions of interest of each of the antibodies may further include calculating the ionic region surface property of each of these regions of interest of each of the antibodies based on the following: calculating the positive ion region surface property of each region of interest of each of the antibodies by aggregating or averaging these patches identified as being associated with each region of interest with a positively charged ionic region (e.g., co-located and / or overlapping). Calculating the negative ion region surface property of each region of interest of each of the antibodies by aggregating or averaging these patches identified as being associated with each region of interest with a negatively charged ionic region (co-located and / or overlapping).

[0112] In yet another example, at least one surface property of a plurality of patches on the surface of each antibody further comprises a hydrophobic surface property. The hydrophobic surface property of each patch comprises an estimated hydrophobic area of ​​each patch. At least one region surface property of each region of interest further comprises a hydrophobic region surface property. The hydrophobic region surface property of each region of interest is calculated by aggregating and averaging the hydrophobic surface properties of the patches identified as being associated with each region of interest (e.g., co-located and / or overlapping).

[0113] In step 316, for each CODV antibody in the plurality of CODV antibodies, the surface properties of the region of interest can be transferred to the PK model generation module 303. Once the surface properties of the plaque have been mapped and combined to the surface properties of each of the regions of interest, multiple linear correlations between the surface properties of the region of interest and the experimental PK values ​​(PK clearance in this case) for all combinations of the regions of interest are performed for the multiple antibodies in the CODV data set. Multiple linear combinations also include additional combinations of surface properties of the regions of interest for different groupings of the regions of interest calculated based on the sum / difference / ratio / score of the combination of two or more surface properties from different regions of interest.

[0114] For example, Figure 3c The regions of interest of the CODV antibody structure 320 are shown. The regions of interest include, but are not limited to, for example, linkers 321 (L1, L2, L3, and L4), VR domains 322 including VR1 and VR2 domains, VH / VL subdomains 324 including VH1, VL1, VH2, VL2, CDR / FW 326 of each of the VH / VL domains 324, CDR / FW 326 including multiple regions 328 VH1-CDR1, VH1-CDR2 to VL2-FW3, and FW4-VL2 regions (e.g., there are 28 VH1 / VL1 / VH2 / VL2-CDR1-3 / FW1-4 regions). Reference Figure 3b , in step 316, based on the average properties of the CODV-Fab arm surface, the average hydrophobic region, ion negatively charged region and ion positively charged region (from, for example, N=50 3D models) located in a separate region of interest (e.g., FW, CDR and joint) are used to calculate a descriptive score or score. In this regard, the sum and ratio of the combination of many different groups of regions of interest (e.g., FW and CDR regions 326 and VH1 / VL1 / VH2 / VL2-CDR1-3 / FW1-4 regions 328) are performed to find the best region of interest grouping. That is, the best region of interest grouping for each of the antibodies is the same grouping, which produces the maximum correlation (e.g., positive or negative correlation) between the descriptive score of the grouping and the experimental PK clearance value of the data set 250. For example, this can include the ratio of ion patches (e.g., patches with ion positive surface properties and / or ion negative surface properties) at a certain position or the total hydrophobic area from a specific area. After performing this task for all possible groupings or combinations of regions of interest, scores (e.g., average surface properties) are generated describing the subdomains (VH1, VL1, VH2, and VL2) and linkers of each VR, as well as their local subregions (a set of CDRs, FWs, or linkers).

[0115] The computer score was used to perform a correlation with the experimental PK clearance data set 250. For the CODV data set 250, a strong exponential relationship was observed in the correlation, indicating that the experimental PK clearance data set 250 should be transformed into its natural logarithm (Ln) counterpart to perform a linear correlation (positive correlation 329a, negative correlation 329b, or no correlation 329c). This may or may not necessarily apply to other data sets for antibodies, etc. In this example, the Pearson correlation coefficient (r), Spearman rank correlation coefficient (r) were also calculated. s ) and the coefficient of determination (r 2 ). Although these types of correlation coefficients are calculated, this is only by way of example and the present invention is not limited thereto, and those skilled in the art will appreciate that any suitable type of correlation coefficient may be applied and / or used. The region of interest group that produces a score or combined score associated with the maximum correlation or maximum absolute correlation (e.g., maximum positive correlation 329a or maximum negative correlation 329b) is selected together with its corresponding linear correlation to form a PK model. The relationship associated with the final combined score can be used by the PK model and can be used as part of the PK region surface property relationship of the PK model to predict or estimate the PK clearance value of an unknown CODV antibody associated with the type of CODV antibody of the data set 250.

[0116] In short, for Figure 2c CODV data set 250, use computer pipeline 310 to analyze the PK properties of CODV-Fab arm.Pipeline 310 needs the sequence of CODV-Fab arm to generate homology model using MOE in steps 311 to 312.Indicate the relevant area name in the three-dimensional structure.Then, in step 313, generate structure set (for example, N=50) by sampling the side chain of the model, and quantify the surface patches of all conformers.Next, in steps 314 and 315, calculate each region of interest such as total hydrophobic surface properties, positively charged surface properties and negatively charged surface properties (see table in step 315) of each FW and CDR in each domain (VH1, VL1, VH2 and VL2) and each joint and / or its combination.The size of the total area of ​​the region of interest indicated by the intensity of gray in the heat map (the darker the color, the larger the total area). In step 316, multiple linear correlations (or nonlinear correlations) are performed for all combinations of the same type of surface properties of the region of interest to find the grouping from the region of interest that has the greatest correlation (positive correlation 329a or negative correlation 329b) with the experimental PK clearance value for each type of surface property (e.g., ionic and hydrophobic surface properties) in all CODV antibody data sets 250. This includes combining the same type of surface properties, such as positive ion ratios and negative ion ratios, by addition, subtraction, multiplication and / or division in various suitable ways.

[0117] Alternatively or additionally, determining the groupings from the region of interest may include determining the ion groupings by combining the corresponding ion region surface properties (e.g., ion positive and ion negative) of each grouping of each of the multiple antibodies for each grouping. This may include based on performing the following steps: aggregating or averaging the positive ion region surface properties of the region of interest in each grouping, and aggregating or averaging the negative ion region surface properties of the region of interest in each grouping. Thus, the combined ion region surface properties of each grouping are determined based on calculating the ratio of the aggregated positive ion region surface properties of the grouping to the sum of the aggregated positive ion region surface properties and the aggregated negative ion region surface properties of the grouping. From all combinations of ion groupings of the region of interest, the ion grouping that produces the maximum correlation (e.g., positive or negative correlation) between the determined combined ion region surface properties of the groupings of each of the multiple antibodies and the corresponding experimental PK clearance value (or any other PK value) of each of the multiple antibodies is selected.

[0118] Alternatively or additionally, determining the groupings from the region of interest further comprises determining, for each grouping, a hydrophobic grouping by combining corresponding hydrophobic region surface properties of each grouping of each of the plurality of antibodies based on aggregating or averaging the corresponding hydrophobic region surface properties in each grouping. Selecting from all combinations of hydrophobic groupings of the region of interest the hydrophobic grouping that produces the greatest correlation (e.g., positive or negative correlation) between the determined combined hydrophobic region surface properties of the grouping of each of the plurality of antibodies and the corresponding experimentally determined PK value of each of the plurality of antibodies.

[0119] The grouping of regions of interest found to have the greatest correlation with clearance for the CODV dataset 250 includes regions of interest VL1-CDR1 and CDR3 for ionic surface properties and regions of interest VR2-FW for hydrophobic surface properties. Two different types of surface properties were identified as being most correlated with PK clearance. These include ionic surface properties (e.g., positive and negative ionic charges) based on the ratio of positive ions in the VL1-CDR1 to CDR3 regions and regions of interest VR2-FW having a hydrophobic surface property. hydrophobic surface properties.

[0120] for Figure 2c For the CODV dataset of 250, all linear correlations scored in silico were analyzed where two different types of surface properties (e.g., ionic and hydrophobic) produced the largest correlations for the group of regions of interest for that type of surface property from all different combinations of regions of interest for that type of surface property. These included the cation ratio in VL1-CDR1 to CDR3 (Equation 1) and the hydrophobic area in VR2-FW (Equation 2).

[0121]

[0122] These relationships associated with the ion score (e.g., Equation 1) and the combined hydrophobicity score (e.g., Equation 2) can be used by a PK model and can be used as part of a PK region surface property relationship to predict or estimate PK clearance values ​​for an unknown CODV antibody associated with the type of CODV antibody in data set 250.

[0123] Figure 4a A set of plots 400 showing the correlation between experimental PK clearance values ​​and ionic properties in two regions of interest (i.e., CDR1 and CDR3 of the VL1 domain in the dataset 250 of CODV Ig-like molecules). Plot 402a shows the positive ion charge area (in terms of The correlation between the negative ionic charge area (measured in units of Å or square angstroms) and the experimental PK clearance value (in vivo clearance ln (mL / h / kg)). Plot 402b shows the negative ionic charge area (measured in units of Å) of the VL1-CDR1 region of the antibodies in dataset 250. The correlation between the positive ion charge area (measured in units of Å or square angstroms) and the experimental PK clearance value (in vivo clearance ln (mL / h / kg)). Plot 404a shows the positive ion charge area (measured in units of Å) of the VL1-CDR3 region of the antibodies in dataset 250. The correlation between the negative ionic charge area (measured in units of Å or square angstroms) and the experimental PK clearance value (in vivo clearance ln (mL / h / kg)). Plot 402b shows the negative ionic charge area (measured in units of Å) of the VL1-CDR3 region of the antibodies in dataset 250. Correlation between the experimental PK clearance value (in vivo clearance ln (mL / h / kg)) and the experimental PK clearance value (in vivo clearance ln (mL / h / kg)). As can be seen in plots 402a and 404a, slow PK clearance molecules show smaller positive areas in CDR1 and CDR3 of the VL1 domain Molecules with rapid PK clearance behave differently. In the plots 402b and 404b of FIG. 404, completely opposite behaviors were observed, indicating that the nature of the ionic properties in the CDRs of the VL1 domain plays an important role in PK clearance. Grouping these scores for this region of interest into a single score explains the positive ion ratios in VL1-CDR1 and CDR3 shown in the plot 406 showing positive ion ratios versus PK clearance (in vivo clearance ln (mL / h / kg)), where the positive ion ratios are obtained according to Equation 1. In all cases, r, r s and r 2The best score in terms of correlation coefficient was the ratio of positive ions in grouping VL1-CDR1 to CDR3, so this relationship can be selected for use in forming a PK region surface property relationship for use in a PK model for predicting or estimating PK clearance of CODVs with unknown PK clearance values.

[0124] Figure 4b A set of plots 410 showing the correlation between experimental PK clearance values ​​and hydrophobic properties in regions of interest (i.e., FW in VH2, VL2, and VR2 in the dataset 250 of CODV Ig-like molecules). Plot 412a shows the hydrophobic area (in terms of The correlation between the hydrophobic area (in units of Å or square angstroms) and the experimental PK clearance value (in vivo clearance ln (mL / h / kg)). Plot 412b shows the hydrophobic area (in units of Å) of the VL2-FW region of the antibodies in dataset 250. The correlation between the hydrophobic area (in units of 100 μg / cm2 or square angstroms) and the experimental PK clearance value (in vivo clearance ln (mL / h / kg)) is shown in Figure 412c for the VR2-FW region of the antibodies in data set 250. The correlation between the hydrophobicity scores (measured in units of square angstroms or square angstroms) and the experimental PK clearance value (in vivo clearance ln (mL / h / kg)). These hydrophobicity scores for the region of interest can be combined to produce the maximum correlation obtained according to Equation 2. In all cases, r, r s and r 2 Slow PK clearance molecules show a smaller hydrophobic area Molecules with rapid PK clearance behave differently. Hydrophobic area in group VR2-FW The highest correlation coefficients from the three hydrophobicity scores are shown.

[0125] Figure 4c A cluster plot 420 is shown, where Figure 4a Drawing 406 and Figure 4b The two best computer scores of plot 412c are plotted against each other. That is, plot 420 is the ratio of positive ions in VL1-CDR1 to CDR3 (e.g., Figure 4a Plot 406) comparing the hydrophobic area in VR2-FW (For example, Figure 4b412c). The marker shapes indicate slow PK clearance molecules (e.g., circles) and fast PK clearance molecules (e.g., triangles). The size of the marker represents the magnitude of the PK clearance value of the antibody (the larger the marker size, the faster the PK clearance). As can be seen, the slow PK clearance molecules tend to be clustered together in the cluster grouping 420b in the lower left corner of the plot 420, while the fast PK clearance molecules tend to be clustered together in the cluster grouping 420a in the upper right area of ​​the plot 420.

[0126] The two selected computer scores describe different regions (VR1 and VR2) and physicochemical properties (ion and hydrophobicity) of CODV-Fab arms, which are relevant to the PK clearance of data set 250. After the hydrophobic area in the control VR2-FW is drawn VL1-CDR1 and the positive ion ratio in CDR3 in the drawing 420, two clusters 420a and 420b rich in slow and fast clearance CODV Ig sample molecules are identified. Two computer scores are used to calculate the combined score (designated as "computer clearance possibility"). The terms "combined score" and "computer clearance possibility" are used interchangeably throughout this disclosure. For constructing the combined score (computer clearance possibility), the weights of the two computer descriptors are equal, and this combined score is obtained according to equation 3.

[0127]

[0128] Assume that one computer score (ion ratio) is a ratio (a value between 0 and 1) and the other computer score (combined hydrophobic average property) is a larger value (between to ), then the scaling percentage of the latter is calculated, which will be in the range between 0 and 1. This helps to obtain a combined score that represents both the ionic score and the hydrophobic score with the same importance (50% each). Figure 4d The correlation between the combined score and the experimental PK clearance values ​​of dataset 250 is shown in FIG.

[0129] Figure 4d is a plot 430 showing the correlation between the experimental PK clearance values ​​of the data set 250 and the combined score obtained from the combination of the two best computer scores: the cation ratio in VL1-CDR1 to CDR3 and the hydrophobic area in VR2-FW The combined score is obtained according to Equation 3. Show r, r s and r 2The slow PK clearance molecule shows a lower combined score (in silico clearance likelihood) than the fast PK clearance molecule. The relationship associated with the combined ion and hydrophobicity score (e.g., Equation 3) can be used by the PK model as a PK region surface property relationship for predicting or estimating the PK clearance value of an unknown CODV antibody associated with the type of CODV antibody of the data set 250. Therefore, according to Figure 1a to Figure 1b The process 100, 110, 120 and Figure 3a and Figure 3b With the system 300 and pipeline 310, a PK model can be generated based on the combined ion and hydrophobicity scores (e.g., Equation 3) and used to predict or estimate the PK clearance value of an unknown CODV antibody, as described in reference Figure 1b and Figure 1c The PK model can be used in technical applications such as, but not limited to, evaluating antibodies with unknown PK values ​​(e.g., clearance) for preliminary screening prior to in vitro wet lab analysis and / or in vivo testing.

[0130] Figure 5 is the ratio of positive ions in VL1-CDR1 to CDR3 versus the hydrophobic area in VR2-FW for two different test CODV molecules not in dataset 250 502a) is a CODV comprising V domains A and B. The second CODV molecule (CODV-BA, 502b) is a CODV comprising the same V domains A and B, but is in different orientations compared with CODV-AB. These are used as test cases, and as known, V domain A has a better PK property (e.g., better PK clearance) than V domain B. In addition, it is also known that V domain B has a larger positive surface area than V domain A on its surface (CDR included). As can be shown, CODV-AB (e.g., CODV 502a) with "good" V domain A as VL1 domain is a CODV-AB showing better computer surface property parameters (including the smaller positive ion ratio in VL1-CDR1 and CDR3 and the smaller hydrophobicity in VR2-FW).

[0131] In contrast, CODV-AB (e.g., CODV 502b) with "bad" V domain B as VL1 domain is a CODV-AB with a very large positive ion ratio in VL1-CDR1 to CDR3 and a larger hydrophobic area in VR2-FW. In addition, in vitro analysis was performed based on FcRn chromatographic data to predict in vivo clearance for these two CODV antibodies 502a and 502b. FcRn chromatographic data describes the retention time associated with PK clearance. In view of this, the longer the retention time in the FcRn data, the worse the expected clearance pattern or PK clearance value of the antibody. The in vitro FcRn chromatographic data of these two CODVs were analyzed, where it was found that the expected "good" CODV-AB antibody 502a had a shorter FcRn retention time than the CODV-BA antibody 502b, which was consistent with the PK model of the CODV molecule.

[0132] These results indicate that the PK model should be relevant to other future unknown CODVs with unknown clearance values. Therefore, there is an alignment between the in silico PK model, the in vitro FcRn wet lab analysis, and the in vivo PK clearance properties. Although the in vivo experimental PK clearance values ​​of these CODV antibodies 502a and 502b are unknown, the PK model predicts that CODV 502a has a better PK clearance value than CODV 502b, and the FcRn chromatogram also indicates this result. This indicates that according to Figure 1a and Figure 3a The PK model pipeline processes 100 and 300 are constructed or formed and applied according to the processes 110, 120 and as described in reference Figures 3a to 4d The described PK model enables accurate prediction of the PK properties of unknown CODV antibodies for initial screening for in vitro wet lab analysis and / or in vivo testing.

[0133] exist Figure 5 The concept shown in Figure 6 The data shown in were experimentally verified. Figure 6 Two plots are shown: plot 602a and plot 603a. Plot 602a shows the ratio of positive ions in VL1-CDR1 to CDR3 versus the hydrophobic area in VR2-FW for two different CODV molecules The values ​​of, each CODV molecule includes V domains C and D, but in different orientations. The two different orientations of the two CODV molecules are CODV-CD (602b) and CODV-DC (602c). From the drawing 602a, it can be shown that the CODV-DC (602c) antibody is predicted to have a better PK clearance value than its counterpart CODV-CD (602b) because it is about the positive ion ratio in VL1-CDR1 and CDR3 and the hydrophobic area in VR2-FW. Both have lower predicted values.

[0134] Drawing 603a shows the combined score (computer clearance probability) of the same two different CODV molecules as in drawing 602a against experimental clearance (in vivo clearance ln (mL / h / kg)). From drawing 603a, it can be shown that the combined score (computer clearance probability) of CODV-DC (603c) is lower than CODV-CD (603b) associated with experimental clearance (in vivo clearance ln (mL / h / kg)); CODV-DC (603c) shows a slower clearance than CODV-CD (603b).

[0135] Figure 7 Contains data set 700 and plot 701. Data set 700 represents a variety of multispecific CODV Ig-like antibodies, including 7 different antibodies from a diverse selection of target specificities not represented in data set 250. Figure 2b The CODV-IgG modality is shown in the framework / structure 220. Plot 701 shows the correlation between the in vivo clearance ln (mL / h / kg) obtained from equation 3 and the combined score (computer clearance probability). Threshold 702 indicates molecules predicted to show a slow clearance curve (<0.5 units in computer clearance probability) or a fast clearance curve (>0.5 units in computer clearance probability). Threshold 703 indicates molecules with slow experimental clearance (<1mL / h / kg or <0 as its natural logarithm counterpart) and fast experimental clearance (>1mL / h / kg or >0 as its natural logarithm counterpart). Display r, r s and r 2 coefficient.

[0136] The predicted PK values ​​for the molecules embedded in the data set 700 and shown in the plot 701 are Figure 1a and Figure 3a The PK model pipeline is generated after processes 100 and 300 and is applied according to processes 110 and 120, and as described in reference Figures 3a to 4dDescribed.When they are predicted, there is no available PK data (e.g., half-life, clearance rate) for any of the 7 molecules embedded in the data set 700.Only based on the PK model pipeline of the present invention, these molecules are selected on purpose from a larger number of available CODV-IgG, to verify the method with the molecule with unknown PK data.This effort is made to identify 4 molecules with slow clearance rate curves and 3 molecules with fast clearance rate curves.In order to determine the clearance rate in vivo by experiment, according to animal use scheme (AUP) and institutional animal care and use committee (IACUC) regulations, 7 kinds of selected CODV-IgG antibodies are expressed in HEK293 cells, two-step purification is carried out to it, and the PK clearance rate in human FcRn transgenic mice (Tg32 hFcRn SCID strain) is characterized.

[0137] Plot 701 shows a very strong correlation between experimental clearance (in vivo clearance ln (mL / h / kg)) and the combined score (in silico clearance likelihood). All seven molecules represented in data set 700 and plotted in plot 701 behave as expected as indicated by Figure 1a and Figure 3a The PK model pipeline processes 100 and 300 are predicted and applied according to processes 110, 120 and as described in reference Figures 3a to 4d As described.PK model and / or process as described herein provide the advantage of efficient and accurate mechanism, and this mechanism significantly reduces the quantity of possible candidate antibody and / or CODV antibody of candidate antibody or CODV antibody final candidate list for wet laboratory analysis.Then, this makes it possible to evaluate and screen a large number of antibodies, wherein, before expensive and / or time-consuming in vitro wet laboratory analysis and / or in vivo test, only those antibodies with required estimation / prediction PK value / characteristic (for example, PK clearance rate) can be preliminarily screened.

[0138] Although the PK model and processes 100, 110, 120 and 310 and the system 300 are described with reference to antibodies or CODV antibodies / CODV data sets 250 / CODV data sets 700, etc., this is by way of example only and the present invention is not limited thereto, and those skilled in the art will appreciate that the PK model and processes 100, 110, 120 and 310 and the system 300 as described herein can be applied to any molecule that can be described in terms of amino acid sequence, such as but not limited to, for example, antibodies, CODV antibodies, enzymes, other proteins, and / or any other suitable protein form, etc. Although PK values ​​have been described herein with respect to generating a PK model, this is by way of example only and the present invention is not limited thereto, and those skilled in the art will appreciate that in addition to PK values ​​such as but not limited to, for example, oligomerization, protein expression, and the like and / or according to application requirements, any other properties of interest may be used.

[0139] Figure 8 is a schematic diagram of a system / apparatus for performing the methods described herein. The system / apparatus shown is an example of a computing device. Those skilled in the art will appreciate that other types of computing devices / systems may alternatively be used to implement the methods described herein, such as a distributed computing system.

[0140] The device (or system) 800 includes one or more processors 802. The one or more processors control the operation of other components of the system / device 800. For example, the one or more processors 802 may include a general-purpose processor. The one or more processors 802 may be a single-core device or a multi-core device. The one or more processors 802 may include a central processing unit (CPU) or a graphics processing unit (GPU). Alternatively, the one or more processors 802 may include dedicated processing hardware, such as a RISC processor or programmable hardware with embedded firmware. Multiple processors may be included.

[0141] The system / apparatus includes a working or volatile memory 804. One or more processors can access the volatile memory 804 to process data and can control the storage of data in the memory. The volatile memory 804 can include any type of RAM, such as static RAM (SRAM), dynamic RAM (DRAM), or it can include flash memory, such as an SD card.

[0142] The system / apparatus includes non-volatile memory 806. The non-volatile memory 806 stores a set of operating instructions 808 in the form of computer-readable instructions for controlling the operation of the processor 802. The non-volatile memory 806 may be any type of memory, such as read-only memory (ROM), flash memory, or magnetic drive memory.

[0143] The one or more processors 802 are configured to execute the operating instructions 808 to cause the system / device to perform the Figures 1a to 7 Any of the methods or processes described. The operating instructions 808 may include code related to the hardware components of the system / device 800 (i.e., drivers), as well as code related to the basic operation of the system / device 800. In general, the one or more processors 802 execute one or more instructions in the operating instructions 808 permanently or semi-permanently stored in the non-volatile memory 806, and use the volatile memory 804 to temporarily store data generated during the execution of the operating instructions 808.

[0144] Embodiments of the methods described herein may be implemented in digital electronic circuitry, integrated circuits, specially designed ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These may include computer program products (such as software stored on, for example, a disk, an optical disk, a memory, a programmable logic device), which include computer readable instructions that, when executed by a computer (such as on Figure 8 When executed by a computer described in the present invention, the computer is caused to perform one or more methods described in this document.

[0145] Any system feature as described herein may also be provided as a method feature, and vice versa. As used herein, means plus function features may alternatively be expressed in terms of their corresponding structures. Specifically, method aspects may be applied to system aspects, and vice versa.

[0146] In addition, any, some and / or all features in one aspect may be applied to any, some and / or all features in any other aspect in any appropriate combination. It should also be understood that the specific combination of various features described and defined in any aspect of the present invention can be independently implemented and / or provided and / or used.

[0147] While several embodiments have been shown and described, it will be appreciated by those skilled in the art that changes may be made in these embodiments without departing from the principles of the present disclosure, the scope of which is defined in the claims.

Claims

1. A computer-implemented method for generating a PK model for predicting pharmacokinetic PK values ​​of an antibody of interest, the method comprising: receiving an input data set of a plurality of antibodies, the input data set comprising data representing an amino acid sequence of each of the antibodies and an experimentally determined PK value for each of the antibodies; calculating one or more surface properties of each of the antibodies based on the corresponding amino acid sequences; calculating one or more region surface properties of one or more regions of interest for each of the antibodies based on the one or more surface properties calculated for each of the antibodies; determining from the one or more regions of interest a grouping that yields a maximum correlation between the corresponding calculated region surface properties of the grouping and the corresponding experimentally determined PK values ​​to establish a PK region surface property relationship for the plurality of antibodies; as well as A PK model for predicting the PK value of the antibody of interest is generated, wherein the PK model is configured to receive one or more input region surface properties of the determined grouping of the antibody of interest, and output a predicted PK value of the antibody of interest by applying the input region surface properties to the PK region surface property relationship.

2. The computer-implemented method of claim 1, wherein the PK model is further configured to calculating one or more surface properties of the antibody of interest based on the amino acid sequence of the antibody of interest; and One or more region surface properties of the defined region grouping of interest for the antibody of interest are calculated based on the one or more surface properties calculated for the antibody of interest, thereby providing the input region surface properties.

3. The computer-implemented method of claim 1 or 2, wherein: The generated PK model is further configured to predict a shortlist of candidate antibodies with desired PK properties for use in in vitro wet lab analysis and / or for use in in vivo trials.

4. The computer-implemented method of claim 3, wherein: For each of these candidate antibodies, the method further comprises the following steps: calculating one or more surface properties of each of the candidate antibodies based on the corresponding amino acid sequence of each of the candidate antibodies; and calculating one or more region surface properties from the determined groupings of regions of interest for each of the candidate antibodies; inputting data representing the calculated regional surface properties of each of the candidate antibodies into the generated PK model for predicting the PK value of each of the candidate antibodies; receiving a predicted PK value for each of the candidate antibodies as an output from the PK model; and When the received predicted PK value of the candidate antibody indicates one or more of the desired PK properties defined for the shortlist of candidate antibodies, adding the candidate antibody to the shortlist; and Data representing a shortlist of candidate antibodies is output for use in at least in vitro wet lab analysis or in vivo testing.

5. A computer-implemented method as claimed in any preceding claim, wherein: Calculating the one or more surface properties of each of the plurality of antibodies includes modeling the three-dimensional molecular structure of each of the antibodies and calculating the distribution of the one or more surface properties on the surface of the modeled molecular structure of each of the antibodies.

6. A computer-implemented method as claimed in any preceding claim, wherein: The calculated one or more surface properties include one or more of positively charged surface area, negatively charged surface area, and hydrophobic surface area.

7. The computer-implemented method of claim 5 or 6, wherein: The surface of the modeled molecular structure of each of the antibodies includes a plurality of patches having one or more surface properties, each patch having a patch area based on the distribution of the surface properties in the modeled molecular structure, wherein the number of patches of each antibody is the same as or different from the number of patches of each other antibody in the plurality of antibodies.

8. The computer-implemented method of claim 7, wherein: Calculating the one or more region surface properties of the one or more regions of interest for each of the antibodies is based on the area of ​​the patch having the one or more surface properties associated with each region of interest.

9. A computer-implemented method as claimed in any preceding claim, wherein: The plurality of antibodies and the antibody of interest are crossover bivariable CODV antibodies.

10. The computer-implemented method of claim 9, wherein: At least one region of interest is selected from the complementary domain region CDR, framework region or linker of the variable heavy VH domain or the variable light VL domain, including CDR1, CDR2, CDR3, FW1, FW2, FW3, FW4 of any one of the VH domain or VL domain of the two variable V domains of these CODV antibodies.

11. The computer-implemented method of claim 10, wherein: Groupings from these regions of interest are CDR1 of VL1, CDR3 of VL1, and FW1 to FW4 of VL2 and VH2 of these CODV antibodies.

12. A computer-implemented method as claimed in any preceding claim, wherein: The PK surface property relationship is a linear PK surface property relationship.

13. A computer-implemented method as claimed in any preceding claim, wherein: Determining a grouping from the one or more regions of interest that produces the greatest correlation between the corresponding calculated region surface property of the grouping and the corresponding experimentally determined PK value comprises: A combined score is calculated from the corresponding calculated zone surface properties of the grouping and the maximum correlation between the combined score and the corresponding experimentally determined PK value is determined to establish a PK zone surface property relationship for the plurality of antibodies.

14. The computer-implemented method of claim 13, wherein: The combined score includes two equally weighted zone surface properties.

15. A device comprising a processor, a memory unit and a communication interface, wherein: The processor is connected to the memory unit and to the communication interface, wherein the processor and the memory are configured to implement the computer-implemented method according to any one of the preceding claims.

16. A computer readable medium comprising data or instruction codes which, when executed on a processor, cause the processor to perform the computer implemented method as claimed in any one of claims 1 to 14.

17. A system comprising: a three-dimensional surface property module configured to receive data representing a plurality of candidate antibodies and calculate a region surface property of a region of interest for each of the candidate antibodies; a pharmacokinetic PK model module configured to receive the calculated regional surface properties corresponding to each of the candidate antibodies for predicting a PK value of each of the candidate antibodies; a PK comparison module configured to compare the predicted PK values ​​of the candidate antibodies with desired PK properties and select those of the plurality of candidate antibodies having PK values ​​that satisfy the one or more desired PK properties; as well as An output module is configured to output a shortlist of candidate antibodies from the selected candidate antibodies based on the comparison for use in in vitro wet lab analysis and / or in vivo testing.

Images