Method and system for developing bio-pharmaceutical

Abstract

To provide a method for developing a bio-pharmaceutical.SOLUTION: There is disclosed a method including: determining experimental data related to at least one monoclonal antibody (mAb); determining calculation derivation data related to at least one mAb, the calculation derivation data including at least one calculation parameter weighted on the basis of the exposure area (ASA) of at least one residue of at least one mAb; determining a plurality of candidate prediction models on the basis of experimental data and calculation derivation data; determining the optimum prediction model from the candidate prediction models; and outputting the optimum prediction model.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] (CROSS REFERENCE TO RELATED APPLICATIONS) This application claims priority to U.S. Provisional Patent Application No. 63 / 108,716, filed November 2, 2020, the entirety of which is incorporated herein by reference. [Background technology]

[0002] Since muromonab, a murine CD3-specific IgG2a monoclonal antibody (mAb) indicated for acute organ transplant rejection, was the first monoclonal antibody to receive FDA approval in 1986, over 64 mAbs have been approved by the FDA. The popularity of this therapeutic platform is evident from the increasing number of ongoing clinical trials each year and the expansion of its use to a wide range of different therapeutic portfolios. Therapeutic mAbs are most commonly administered via three routes of administration: intravenous (IV), intramuscular (IM), and subcutaneous (SC) injections, with the choice based on a variety of contributing factors including their safety, efficacy, patient satisfaction, and pharmacoeconomics. IV administration can usually be delivered in high doses that are controllable in the clinic and therefore is usually more costly for patients and clinicians. The transition of most immunoglobulins from the IV route to the SC route of administration may reduce overall healthcare costs by allowing patients to self-administer at home or by healthcare professionals to administer faster in the clinic, reducing prolonged hospital stays and easing the economic burden on the healthcare system by improving the overall quality of care. Despite offering many advantages over IV administration, the SC route of administration presents several significant challenges for drug development and drug administration. The main drawbacks of SC administration are the natural resistance to extracellular matrix and volume limitations, which require highly concentrated antibody solutions (≥150 mg / mL) to be administered in limited injection volumes (~2–3 mL) for optimal PK / PD outcomes and user convenience.

[0003] Highly concentrated antibody solutions are difficult to develop in the pharmaceutical field, since high protein concentrations can lead to significant technical challenges, such as high solution viscosity and protein aggregation rates. Highly viscous antibodies also lead to difficulties related to manufacturing processes and drug delivery. Protein aggregation can lead to reduced antibody activity and greater immunogenicity potential, which may affect the pharmacokinetics and safety of the protein. Therefore, suitable formulations should be developed for highly concentrated therapeutic antibodies to stabilize both the colloidal and conformational structures, thereby reducing viscosity and aggregation tendency, and ensuring both an acceptable shelf life and compatibility with manufacturing processes.

[0004] Traditional approaches to developing high concentration protein formulations generally involve empirical methods utilizing a wide range of biophysical and analytical techniques. For example, solution viscosity is measured by direct viscometry and is based on the osmotic second virial coefficient (B 22 ) measurement, and the diffusion interaction parameter (K D ) measurements. Such tools, e.g., B 22 The measurement requires significant amounts of material and is also fairly labor intensive. Protein aggregation and association are thought to be the result of an interplay between conformational stability (i.e., macroscopic and microscopic perturbations in the protein structure) and colloidal stability (i.e., native intermolecular interactions).

[0005] A variety of well-established biophysical and analytical tools and approaches are routinely used to assess overall protein stability and predict the individual contributions of conformational and colloidal stability. These stabilities can be measured and predicted by a variety of established techniques that allow for the quantification of thermal stress, agitation stress, and freeze-thaw stress. Stability prediction approaches include thermal stress stability studies, thermal denaturation temperatures (T m ), chemical denaturation temperature, coagulation temperature (T agg ), cloud point temperature (T cloudThese techniques include the measurement of surface hydrophobicity, direct surface hydrophobicity measurement using hydrophobic interaction chromatography, zeta potential, and higher order structure estimation. All of these techniques require physical materials and some are cumbersome and time consuming.

[0006] Moreover, experimentally developed predictive models have failed in most cases. Most of the above mentioned techniques for measuring and predicting viscosity and aggregation rates are costly, time consuming, and require physical materials. Therefore, the development of novel experimental and / or computational tools to predict viscosity values ​​and aggregation propensity or to rank antibodies based on their developability is essential for rapid screening of many mAb candidates during early formulation development and drug discovery.

[0007] Physical properties derived from antibody sequences, homology models, and molecular dynamics (MD) simulations of individual antibody molecules can be used as parameters to develop predictive models or ranking schemes. These models can be used to predict viscosity, conformational stability, colloidal stability, and manufacturability. These rapid and materials-free tools also enable molecular insight into antibody molecules and their interactions.

[0008] Sharma et al. developed a predictive model of viscosity based on the variable domain Fv of 14 IgG1 homology models. Physical parameters including but not limited to charge and hydrophobicity were correlated to the measured viscosity values ​​in a principal component regression model. Agrawal et al. also developed a viscosity scoring function to rank IgG1 antibodies based on the partial charges of surface-exposed residues of the Fv region in the homology model and proposed a threshold value to distinguish highly viscous antibodies from other antibodies.

[0009] predicted the concentration-dependent viscosity curves of antibody solutions based on electrostatic and hydrophobic descriptors derived from full-length homology models of 16 IgG1, IgG2, and IgG4 antibodies. The hydrophobic surface area of ​​the full-length antibodies and the charges on the Fv and hinge regions were used to predict the slope of the linearized concentration-dependent viscosity curves.

[0010] Furthermore, various methods have been developed by researchers in the field to predict aggregation-prone regions of peptides and therapeutic proteins. The TANGO statistical mechanics algorithm was developed based on the physicochemical principles of beta-sheet formation to predict sequence-based aggregation. Waltz, a web-based tool, was designed to identify amyloidogenic regions in protein sequences using position-specific scoring matrices. Chennamsetty et al. developed a method to predict aggregation-prone regions based on the dynamic exposure of hydrophobic patches obtained from atomistic simulations of antibodies. A comprehensive list of computational methods developed to predict aggregation and aggregation-prone regions of therapeutic proteins can be found in published book chapters and review books.

[0011] Overall, there is a need for more robust and predictive models for viscosity and aggregation propensity to facilitate drug development. Summary of the Invention

[0012] A method is described that includes determining experimental data related to one or more monoclonal antibodies (mAbs); determining computationally derived data related to the one or more mAbs, where the computationally derived data includes one or more computational parameters weighted based on an exposed area (ASA) of one or more residues of the one or more mAbs; determining a plurality of candidate predictive models based on the experimental data and the computationally derived data; determining an optimal predictive model from the plurality of candidate predictive models; and outputting the optimal predictive model.

[0013] Also described is a method that includes receiving computationally derived data associated with a monoclonal antibody (mAb), providing the computationally derived data to a predictive model, and determining a viscosity score associated with the mAb based on the predictive model.

[0014] Also described is a method that includes receiving computationally derived data associated with a monoclonal antibody (mAb), providing the computationally derived data to a predictive model, and determining an agglutination score associated with the mAb based on the predictive model.

[0015] Additional advantages of the disclosed method and compositions will be set forth in part in the description which follows and in part will be understood from the description or may be learned by practice of the disclosed method and compositions. The advantages of the disclosed method and compositions will be realized and attained by means of the elements and combinations particularly pointed out in the appended claims. It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention as claimed.

[0016] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate certain embodiments of the disclosed methods and compositions and, together with the description, serve to explain the principles of the disclosed methods and compositions. [Brief description of the drawings]

[0017] [Figure 1] FIG. 1 is a flow chart illustrating an example for generating a predictive model to aid in screening and / or selection for a treatment. [Diagram 2] FIG. 2 is an example of a block diagram for generating a predictive model. [Diagram 3] FIG. 3 is a flow chart illustrating an example of a training method. [Figure 4]FIG. 4 is a diagram of an exemplary process flow for determining whether a nucleotide sequence is a promoter using a machine learning based classifier. [Diagram 5] Figure 5 shows examples of diffusion interaction parameters (KD) calculated based on fitting a line to the diffusion coefficient at various protein concentrations. The graph shows how the KD was calculated for mAb4 as an example. [Figure 6A] Figures 6A and 6B are tables showing the calculated parameters for the 16 full antibody models used in this study. These physical properties were obtained from the full antibody homology models used to develop predictive models for protein solution viscosity. ZVL, ZVH, ZCL, ZCH1, ZHinge, ZCH2, and ZCH3 are the net charges for the VL, VH, CL, CH1, hinge, CH2, and CH3 regions, respectively; ZmAb is the total antibody charge; Z*VL, Z*VH, Z*CL, Z*CH1, Z*Hinge, Z*CH2, and Z*CH3 are the net solvent accessible area (SAS) adjusted charges for the VL, VH, CL, CH1, hinge, CH2, and CH3 regions, respectively; HI is the hydrophobicity index; DmAb is the averaged total dipole moment of the antibody; PISequence and PIStructure are the sequence-based and structure-based isoelectric points, respectively; and AP is the aggregation propensity predicted by Chennamsetty. [Figure 6B] Same as above. [Figure 7]Figure 7 is a table showing the calculated parameters of the Fab models. In this study, the calculated parameters of 14 Fab models were used. These physical properties were obtained from the Fab homology model and molecular dynamics simulations used to develop a predictive model for aggregation propensity. ZVL, ZVH, ZCL, ZCH1 are the net charges for the VL, VH, CL, and CH1 regions, respectively; ZFab is the total Fab region charge; Z*VL, Z*VH, Z*CL, Z*CH1 are the net solvent accessible area (SAS) adjusted charges for the VL, VH, CL, and CH1 regions, respectively; HI is the hydrophobicity index; DmAb is the averaged total dipole moment of the Fab regions; PISequence and PIStructure are the sequence-based and structure-based isoelectric points, respectively; AP is the aggregation propensity predicted by Chennamsetty; and RMSD is the averaged root mean square deviation (Å) obtained from molecular dynamics simulations of the Fab regions. [Figure 8] Figure 8 shows the broad distribution of protein solution viscosity values ​​measured for the 16 mAbs used in this study. IgG1 and IgG4 candidates are shown in grey and black, respectively. According to our data set, IgG1 antibodies tend to show lower viscosity values ​​compared to IgG4 candidates. [Figure 9] Figures 9A, 9B, and 9C show examples of measurements for the 15 mAbs used in this study. (a) Osmotic second virial coefficient (B22), (b) Diffusion interaction parameter (KD), and (c) Strong correlation between B22 and KD values ​​for the current data set, as observed by a high correlation coefficient (R). MAb1 was excluded from these plots because there was no material for KD measurement. IgG1 and IgG4 candidates are colored gray and black, respectively. [Figure 10]Figures 10A and 10B show examples of correlations of measurements to protein solution viscosity: a) the osmotic second virial coefficient (B22), and (b) the diffusion interaction parameter (KD). B22 values ​​were measured for the 16 mAbs used in this study, and KD values ​​were measured for 15 of them because there was not enough material for mAb1. The linear correlation coefficient (R) and the regression line are shown on each graph. [Figure 11A] Figures 11A, 11B, and 11C show the linear relationships between the experimental viscosity values ​​and the calculated parameters. The linear fit equations and correlation coefficients (R) are shown for each plot. ZVL, ZVH, ZCL, ZCH1, ZHinge, ZCH2, and ZCH3 are the net charges for the VL, VH, CL, CH1, hinge, CH2, and CH3 regions, respectively; ZmAb is the total antibody charge; Z*VL, Z*VH, Z*CL, Z*CH1, Z*Hinge, Z*CH2, and Z*CH3 are the net solvent accessible area (SAS) adjusted charges for the VL, VH, CL, CH1, hinge, CH2, and CH3 regions, respectively; HI is the hydrophobicity index; DmAb is the averaged total dipole moment of the antibody; PISequence and PIStructure are the sequence-based and structure-based isoelectric points, respectively; and AP is the aggregation propensity predicted by Chennamsetty. [Figure 11B] Same as above. [Figure 11C] Same as above. [Figure 12] Figure 12 shows an example of a linear regression line between the calculated predicted viscosity score (PVS) and the measured viscosity values. The correlation coefficient (R) and the squared correlation coefficient (R2) are shown on the graph. The area between the dashed lines indicates the 95% confidence interval. [Figure 13]Figure 13 shows a representative example of a size exclusion chromatography (SEC) signal for one of the antibodies mAb3 used in this study, showing incubation at 40°C and 75% relative humidity over a period of 0 and 28 days. An increase in high molecular weight (HMW) species formation is observed as a result of aggregation. Data for days 0 and 28 are shown as dotted red and solid black lines, respectively. [Figure 14] Figure 14 shows the relative percentage of high molecular weight (HMW) species formation, %ΔHMW, measured by size exclusion chromatography (SEC) for the 14 mAbs used in this study at 7, 14, and 28 days compared to day 0. Samples were incubated at 40°C and 75% relative humidity for 7, 14, and 28 days. MAb9 and mAb16 were excluded due to limited material availability. [Figure 15] Figure 15 shows the daily rate of high molecular weight (HMW) species formation, %ΔHMW / day, for the 14 mAbs used in this study, calculated based on the %ΔHMW of the 28 day data points divided by 28. IgG1 and IgG4 are colored grey and black, respectively. [Figure 16A] Figures 16A, 16B, and 16C show the root mean square deviation (RMSD) of the conformational structures relative to the initial structures of the 14 mAbs used in this study. These conformations were obtained from three 2.0 ns molecular dynamics simulations of the Fab region of each antibody. The first, second, and third simulations of each mAb are colored black, red, and blue, respectively. [Figure 16B] Same as above. [Figure 16C] Same as above. [Figure 17]Figure 17 shows examples of averaged root mean square deviations (RMSD) of backbone atoms of Fab models from the initial structures of 14 mAbs used in the current study to develop predictive models of aggregation. The RMSD values ​​can be used as a descriptor of conformational stability to distinguish antibodies from each other. The RMSD values ​​for each mAb are the average of three molecular dynamics (MD) simulations. MAb9 and mAb16 were excluded due to limited material availability. [Figure 18A] Figures 18A, 18B, and 18C show the averaged root mean square deviation (RMSD) for three 2.0 ns molecular dynamics simulations of the Fab region of each mAb. The RMSD for each simulation was calculated for each conformation relative to the initial structure. [Figure 18B] Same as above. [Figure 18C] Same as above. [Figure 19A] Figures 19A, 19B, and 19C show the linear relationship between the measured daily high molecular weight (HMW) species formation rate, %ΔHMW / day, and calculated parameters, including the root mean square deviation (RMSD), obtained from molecular dynamics simulations. [Figure 19B] Same as above. [Figure 19C] Same as above. [Figure 20] Figure 20 shows an example of a linear regression line between the calculated predicted aggregation score (PAS) and the measured daily high molecular weight (HMW) formation rate. The correlation coefficient (R) and the squared correlation coefficient (R2) are shown on the graph. The area between the dashed lines indicates the 95% confidence interval. [Figure 21] FIG. 21 shows the correlation between predicted viscosity scores and validation experimental data. [Figure 22] FIG. 22 shows the correlation between predicted aggregation scores and validation experimental data. [Diagram 23] FIG. 23 illustrates an exemplary operating environment. [Figure 24] FIG. 24 illustrates an exemplary method. [Diagram 25] FIG. 25 illustrates an exemplary method. [Figure 26]FIG. 26 illustrates an exemplary method. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0018] The disclosed methods and compositions can be readily understood by reference to the following detailed description of the specific embodiments and examples contained therein, as well as the drawings and accompanying description.

[0019] It is to be understood that the methods and compositions of the present disclosure are not limited to the specific methodology, protocols, and reagents described, since these may vary. It is also to be understood that the terminology used herein is for the purpose of describing particular embodiments only, and is not intended to limit the scope of the present invention, which is limited solely by the appended claims.

[0020] It should be noted that as used in this specification and the appended claims, the singular forms "a," "an," and "the" include plural references unless the context clearly dictates otherwise. Thus, for example, a reference to "an antibody" includes a plurality of such antibodies, a reference to "the antibody" is a reference to one or more antibodies and equivalents thereof known to those skilled in the art, and so forth.

[0021] As used herein, the term "antibody" refers to an entire antibody. An antibody is a glycoprotein that contains at least two heavy (H) chains and two light (L) chains interconnected by disulfide bonds. Each heavy chain is composed of a heavy chain variable region (abbreviated herein as VH) and a heavy chain constant region. The heavy chain constant region is composed of three subdomains: CH1, CH2, and CH3. Each light chain is composed of a light chain variable region (abbreviated herein as VL) and a light chain constant region. The light chain constant region is composed of one subdomain, CL. The VH and VL regions can be further subdivided into regions of hypervariability, termed complementarity determining regions (CDRs), interspersed with more conserved regions, termed framework regions (FRs). Each VH and VL is composed of three CDRs and four FRs arranged from amino terminus to carboxy terminus in the following order: FR1, CDR1, FR2, CDR2, FR3, CDR3, FR4. The variable regions of the heavy and light chains contain the binding domains that interact with antigens. The constant regions of the antibody can mediate the binding of the immunoglobulin to host tissues or factors, including various cells of the immune system (e.g., effector cells) and the first component (Clq) of the classical complement system. In some embodiments, the antibody may be a chimeric antibody, a monoclonal antibody, and / or a humanized antibody.

[0022] An antibody fragment can refer to any smaller portion of a whole antibody. Antibody fragments can be described in terms of proteolytic fragments, including but not limited to Fv (fragment variable region), Fab (fragment antibody binding region), Fab', and F(ab')2 (fragment antibody binding region plus part of the hinge region) fragments. Such fragments may be prepared by standard methods (see, for example, Coligan et al. Current Protocols in Immunology, John Wiley & Sons, 1991-1997, incorporated herein by reference). An antibody may contain at least three proteolytic fragments (i.e., fragments produced by cleavage with papain): two Fab fragments (herein designated as "Fab heavy chain domains"), each containing a light chain domain and a heavy chain domain, and one Fc fragment containing two Fc domains. Each light chain domain contains a VL subdomain and a CL subdomain, each Fab heavy chain domain contains a VH subdomain and a CH1 subdomain, and each Fc domain contains a CH2 subdomain and a CH3 subdomain. In some embodiments, the antibody fragment may be a chimeric antibody fragment, a monoclonal antibody fragment, and / or a humanized antibody fragment.

[0023] As used herein, the term "monoclonal antibody" or "monoclonal antibody fragment" refers to an antibody or antibody fragment obtained from a single clonal population of immunoglobulins that binds to the same epitope of an antigen. Monoclonal antibodies have the same Ig gene rearrangement and therefore exhibit identical binding specificity. Methods for preparing monoclonal antibodies are known in the art.

[0024] As used herein, a "humanized monoclonal antibody" or "humanized monoclonal antibody fragment" may refer to a monoclonal antibody or fragment thereof having at least a human constant region and an antigen-binding region, e.g., one, two, or three CDRs, derived from a non-human species. A humanized antibody or fragment thereof specifically recognizes an antigen of interest but does not elicit an immune response in humans against the antibody itself.

[0025] As used herein, the term "chimeric antibody" or "chimeric antibody fragment" refers to a monoclonal antibody or fragment thereof that contains a variable region from one source (e.g., species) and at least a portion of a constant region derived from a different source. In some embodiments, a chimeric antibody contains a murine variable region and a human constant region.

[0026] Throughout this specification and the claims, the word "comprise" and variations of this word, such as "comprising" and "comprises," mean "including, but not limited to," and are not intended to exclude, for example, other additional things, components, integers, or steps. In particular, in methods described as including one or more steps or operations, each step is specifically contemplated to include what is recited (unless that step includes a limiting term such as "consisting of"), meaning that each step is not intended to exclude, for example, other additional things, components, or steps not recited in the step.

[0027] During drug discovery and early development, most drug candidates are initially screened and selected based on affinity and functionality. However, there are other properties and attributes that need to be considered in biopharmaceutical development. For example, protein yield, viscosity, aggregation, chemical stability (e.g., susceptibility to degradation by oxidation, deamidation), compatibility, and immunogenicity should form part of a comprehensive developability risk assessment. The concept of developability is used to define the suitability of a drug candidate (e.g., an antibody) to be developed as a therapeutic / drug. Once an antibody is identified and developed as a drug, it can be administered to patients.

[0028] A method for predicting antibody viscosity is disclosed. This viscosity prediction tool can be used early in the drug development process, allowing for ranking and selection of lead antibodies with reduced risk of being sticky.

[0029] Antibody drugs require highly concentrated formulations, which can result in solutions that are difficult to handle and highly viscous both in the manufacturing process and for injection into the end user. Antibody viscosity is often only discovered in the later stages of drug development, after significant research and development investments. The methods and systems described herein can predict antibody viscosity at an earlier stage in the drug development process, allowing for prioritization of antibody candidates with lower viscosity. Although the viscosity prediction tools relate to the analysis of antibodies, it should be understood that these techniques can also be applied to other proteins. The protein can be, but is not limited to, a clinical candidate.

[0030] A method to predict antibody aggregation is disclosed. This aggregation prediction tool can be used early in the drug development process, allowing for ranking and selection of lead antibodies with reduced aggregation risk. Protein aggregation, a commonly occurring issue during biopharmaceutical development, can occur at different stages of the manufacturing and development process, including fermentation, purification, formulation, fill and formulation finalization, and storage. Aggregation can impact not only the manufacturing process, but also the target product profile, product efficacy, delivery, and importantly, patient safety. Protein aggregates have been reported to contribute to cases of immune reactions in patients.

[0031] These aggregates may manifest as reversible oligomers, sub- or macroscopic particles, or precipitates.The protein aggregation process is determined by many factors, including the amino acid composition and sequence, environmental factors such as pH, concentration, buffers / excipients and shear forces during the process used to produce the protein, as well as the final formulation and storage conditions.

[0032] In some embodiments, the aggregation prediction can be used in combination with other computational prediction tools to screen and select antibodies of interest. For example, the disclosed aggregation prediction model can be combined with the disclosed viscosity prediction model, or with known immunogenicity or degradation prediction tools. The combination of these tools can select one or more antibodies with reduced risk of aggregation, viscosity, degradation, and / or immunogenicity for in vitro expression and characterization.

[0033] In one embodiment shown in Figure 1, a method 100 is described for generating a predictive model to aid in therapeutic screening, ranking, and / or selection. At 110, experimental parameters may be determined. The experimental parameters may relate to, for example, protein yield, viscosity, aggregation, chemical stability (e.g., susceptibility to degradation by oxidation, deamidation), compoundability, and / or immunogenicity.

[0034] The experimental parameters may be determined from experimental data. The experimental data may be, for example, data generated by measurements, test methods, experimental designs, and / or quasi-experimental designs. In clinical research, the data generated are the results of clinical trials. The experimental data may be qualitative or quantitative, each suitable for different investigations. The experimental data may include values ​​of the experimental parameters obtained by performing one or more experiments related to the antibody.

[0035] In an embodiment, experimental parameters related to viscosity may be determined. In some aspects, techniques for measuring viscosity measure how a sample responds to flow rate, velocity, and time. For example, a capillary viscometer may be used to measure the time it takes for a sample to pass through a tube. Similar to using a capillary viscometer, the Zahn cup method may be used in which a small hole is placed in the bottom of a cup, and the time it takes for the sample to pass through the hole is measured. A falling ball viscometer technique may also be used to measure viscosity, where a ball of known density is dropped into the sample, and the time it takes for the ball to fall to a specified point is recorded. In some aspects, a vibrational viscometer is used to measure the damping of a vibrating electromechanical resonator immersed in the sample. A rotational viscometer technique may be used, which measures the torque required to rotate an object in the sample as a function of the viscosity of the sample.

[0036] In one embodiment, experimental parameters related to antibody aggregation may be determined. Experimental parameters related to antibody aggregation may be performed using any known protein aggregation technique. For example, biochemical assays for measuring aggregation include, but are not limited to, ultracentrifugation, size exclusion chromatography, gel electrophoresis, dynamic light scattering or turbidity measurements. Many of these techniques take into account the size difference between protein monomers and aggregates. Fluorescence-based assays may be used, in which a fluorophore increases its fluorescence yield in the presence of protein aggregates.

[0037] In one embodiment, experimental parameters related to protein yield can be performed using techniques known in the art. Protein concentration is similar to protein yield, but establishes the amount of protein in a specific volume of solution. Protein concentration is most often determined using a spectrophotometer. Once protein concentration is determined, protein yield can be determined. Thus, if a sample has a protein concentration of 5mg / ml, the total protein yield is 500mg when the protein yield is 100ml.

[0038] In one embodiment, the experimental parameters related to antibody-antigen docking can be carried out using known techniques. The "standard" for obtaining this data is by experimentally determining the 3D structure of the antibody-antigen complex using X-ray crystallography. Other structural methods such as cryo-electron microscopy (cryoEM) or nuclear magnetic resonance (NMR) can also be used, although the latter is difficult due to the size of the complex. These experimental data showing the possible binding between an antibody and its antigen may result in different conformational changes that may occur upon binding.

[0039] In one embodiment, experimental parameters related to immunogenicity may be determined. The immunogenicity of therapeutic antibodies may cause adverse side effects. Immunogenicity may be experimentally determined using animal experiments. Antibodies may be administered to animals (such as mice or rabbits), and then at different time points, serum from the animals may be tested for immune responses to the antibodies (especially T-cell and B-cell responses). In most cases, the lower the immunogenicity, the better the therapeutic antibody choice. In some aspects, the immunogenicity of an antibody may be altered by humanizing the antibody.

[0040] In one embodiment, experimental parameters related to chemical stability may be determined. Chemical stability can be an important attribute of therapeutic proteins, especially antibodies. In most cases, the more likely an antibody is to degrade, the less desirable it is as a therapeutic agent. The most common method for experimentally determining chemical stability is to use gel electrophoresis. Pulse-chase assays can also be used. pH, temperature, and proteases are all factors in chemical stability. Thus, even slight formulation changes can affect chemical stability.

[0041] At 120, computational parameters may be determined. The computational parameters may be determined by computational analysis and / or simulation. The computational parameters may be determined from computationally derived data. The computationally derived data may be data generated by, for example, sequence analysis, antibody numbering, full FV region modeling, Ab-specific side chain prediction, antibody-specific loop prediction, side chain prediction, first-principles loop prediction, CDR canonical structure prediction, VH / VL orientation, paratope prediction, protein contact prediction, Ab-specific epitope prediction, Ab-specific docking, non-specific docking, structure prediction, homology modeling, protein-protein docking simulation, molecular dynamics simulation, etc. The experimental data may include values ​​of experimental parameters obtained by performing computational analysis related to the antibody.

[0042] In one embodiment, the computational parameters may be determined through antibody numbering. The antibody sequence may be mapped onto a standardized reference framework. The raw nucleotide sequences of the variable regions may be translated into amino acids by aligning them to the germline sequence, thereby identifying the V, D, and J regions. This may be accomplished by programs such as IgBLAST or IMGT V-Ques and multiple other tools aimed at processing raw antibody data. The similarity between antibody amino acid sequences further allows the generation of a standardized reference framework, or numbering scheme, giving an identifier to each variable region amino acid. The numbering scheme describes the context of each position within the structure of the antibody and allows for the rapid delineation of the CDRs and framework regions. Antibody numbering may be the first step in computational antibody analysis, such as homology modeling.

[0043] In one embodiment, the computational parameters may be determined through antibody modeling. Structural antibody modeling involves generating a 3D structure from an antibody sequence based on existing knowledge of antibody structure in particular, and protein structure in general. The high degree of antibody sequence and structure conservation in the framework regions and the five canonical loops results in an overall high accuracy of antibody homology modeling. Antibody modeling generally involves the selection of a suitable framework template that can accommodate the CDR loops. This can be achieved by searching for close sequence matches with the H and L chains in the available databases. The relative orientation of the VH and VL domains is determined, which affects the shape of the paratope. The CDR loops are then modeled. An antibody-specific knowledge-based approach can be used to predict the CDR loops according to the template. In the absence of a suitable template, as is often the case with CDRH3, a more computationally expensive first-principles approach can be used to generate a large set of novel loops and select the best loop model. Side chains are then built and refined. A protein-general and / or antibody-focused approach may be employed. The final antibody model can be further refined by optimizing the energy loading of the molecule. Various modeling tools may be used, such as Biovia from Accelrys (https: / / www.3dsbiovia.com / ), SmrtMolAntibody from Macromoltek (https: / / www.macromoltek.com / ), MOE from CCG (https: / / www.chemcomp.com / ), and BioLuminate from Schrodinger Inc. (https: / / www.schrodinger.com / products / bioluminate). The modeling tools provide accurate, on average, root mean square deviation (RMSD) measurements of antibody F. VAn entire model can be generated, with the most challenging region being CDRH3, which is modeled to an RMSD of >5 Å for some targets. Although such results do not usually match the accuracy of experimentally derived structures, models with RMSDs of 1.0 Å can be used as a steep surrogate for delineating the structural features of the molecule. Model structures can be used at selected surface-exposed paratope residues for mutations, or to characterize binding to cognate epitopes. Accurate structural information can be used to assess various developability metrics, such as hydrophobicity, which depends on accurate models for the molecular surfaces of the paratope and epitope.

[0044] In one embodiment, computational parameters related to residue charge may be determined via antibody homology modeling. A homology model of the complete antibody and / or Fab (antigen-binding fragment) may be constructed via modeling software using the Protein Data Bank (PDB) crystal structure as a template. In one embodiment, a complete antibody and / or Fab homology model may be constructed to determine computational parameters of protein viscosity and / or protein aggregation tendency. As described herein, the complete antibody and / or Fab homology model may be used in molecular dynamics simulations to determine the computational parameters. The energy of the antibody structure may be determined based on the homology model and then minimized by geometry optimization. The antibody structure may be protonated, followed by the determination of computational parameters such as charge and average dipole moment for the residues.

[0045] An antibody (Z in the full antibody model) containing both the light and heavy chain variable and constant regions. VL , Z VH , Z CL , Z CH1 , Z Hinge , Z CH2 , Z CH3 , Z Total , and Z for the Fab model. VL , Z VH , Z CL , ZCH1 , Z Total ) may be determined as a calculation parameter. In one embodiment, the net charge of each residue may be adjusted by considering the relative solvent accessible area (SAS) of that residue in the homology model of the corresponding antibody. In one embodiment, a built-in algorithm in Discovery Studio software may be used to determine the total exposed surface area of ​​each amino acid. In this approach, depending on the model being used, the charge for each residue may be multiplied by a weighting factor calculated using the SAS of the residue relative to the total SAS of either the full antibody or the Fab. For example, in the variable light chain, the adjusted charge for each residue may be calculated using Equation 3, and the total SAS adjusted charge for this region may be calculated using Equation 4. These SAS adjusted charges are calculated using Equation 5 for the full model, Z * VL , Z * VH , Z * CL , Z * CH1 , Z * Hinge , Z * CH2 , Z * CH3 , Z * Total , and Z for the Fab model. * VL , Z * VH , Z * CL , Z * CH1 , Z * Total It may be labeled as:

[0046]

number

[0047] where i = any residue in the variable light chain (VL), and n = the number of residues in the full model or Fab model of the specified antibody.

[0048]

number

[0049] where m = the number of residues in the variable light chain (VL). In one embodiment, the full antibody and / or Fab homology model may be used to determine the hydrophobicity index (HI) as a calculation parameter. The HI of a variable fragment (Fv) is:

[0050]

number

[0051] where i represents a hydrophobic amino acid, e.g., A, C, F, G, I, L, M, P, V, W, Y, j represents a hydrophilic amino acid, e.g., D, E, H, K, N, Q, R, S, T, n is the number of each amino acid, and E is the Eisenberg scale value of each amino acid. In one embodiment, the full antibody and / or Fab homology model may be used to determine the average dipole moment (HI) as a calculation parameter. The average dipole moment of the full model and the Fab model can be determined from the protonated structure.

[0052] In one embodiment, the full antibody homology model, the Fab homology model, and / or the antibody sequence data may be used to determine the isoelectric point (pI). In one embodiment, the full antibody and / or the Fab homology model may be used to determine the aggregation propensity (AP) score for each atom. The AP score may be determined, for example, based on the CHARMM force field and SAS patch of exposed hydrophobic residues in a radius of 10 Å. The total aggregation score for each antibody may be determined as the sum of the aggregation scores of all residues in either the full antibody homology model or the Fab homology model. Examples of calculation parameters that may be determined based on the antibody homology model and / or the Fab homology model are shown in Table 1.

[0053] [Table 1]

[0054] In one embodiment, the computational parameters may be determined based on molecular dynamics (MD) simulations. MD simulations may be utilized to include conformational changes in the Fab region that are related to aggregation tendency. Atoms of the Fab structure may be assigned force field parameters by structurally matching each residue to its template. These structures may be explicitly solvated within a truncated octahedral box of TIP3P water molecules. The counterion Na + and Cl - may be added to an explicitly solvated system to neutralize the system. In each simulation, the energy of the system may be minimized using a steepest descent algorithm, and then further minimized using an Adopted Basis Newton-Raphson (ABNR) minimization method to remove large distortions in the system. The system may be gradually heated under constant volume (NVT) and simulated at constant temperature and pressure. The Particle Mesh Ewald (PME) method may be used to determine long-range electrostatics using a cutoff distance for van der Waals interactions. The SHAKE algorithm may be used in each simulation to constrain the bond lengths for all hydrogen atoms. Simulations may be performed for each system, differing only in the initial distribution of velocities, to allow scrutiny of the results reproducibility. The trajectory of each simulation, the atomic coordinates varying in time, may be captured. In one embodiment, the backbone root mean square deviation (RMSD) of the conformational structure relative to the initial structure after rigid body alignment in each simulation may therefore be determined as a computational parameter that is a descriptor of conformational stability.

[0055] In one embodiment, the computational parameters may be determined through interface prediction and antibody-antigen docking. Computational methods may be employed to predict the antibody-antigen contact surface. Computational methods may predict, for example, the paratope, epitope, or the entire antibody-antigen complex. Approximately half of the 40-50 residues in the CDRs directly contact the antigen and form the paratope. Statistical methods such as Antibody i-Patch assign a score to each residue for the residue's propensity to be part of the paratope, with high-scoring residues providing potential candidates for mutagenesis. Because not all paratope residues are constrained by the CDRs, positions within the framework regions that may contribute to antigen recognition can be computationally identified.

[0056] Computational methods for epitope prediction can be divided into linear epitope predictors, which focus on identifying continuous stretches of primary amino acid sequence, and conformational epitope predictors, which aim to identify the 3D configuration of the epitope. Although paratope and epitope prediction can provide useful information about antibody-antigen recognition, these methods do not provide information about the specific interactions involved in antibody-antigen binding. This problem is addressed by antibody-antigen docking, a specialized application of the broader field of molecular docking. Molecular docking predicts biological complexes starting from unbound proteins. It typically involves two steps: a sampling step, during which thousands of possible conformations of the complex are generated, and a scoring step, which ranks the conformations according to a specific scoring function to identify models that are closer to the native conformation.

[0057] In one embodiment, computational parameters may be determined through evaluation of the "humanization" of therapeutics via sequence analysis. The majority of antibodies currently developed are discovered by immunization of animals. Molecules generated in animals, such as mice, carry the risk of eliciting an immunological response in humans in the form of anti-drug antibodies (ADA). To avoid such problems, animal-derived antibodies undergo a process called humanization. During this process, CDRs from (typically) mouse-derived antibodies are grafted onto a human framework, or alternatively, the mouse-derived framework is engineered to resemble the human framework. Traditionally, humanization involves comparing an animal-derived sequence to approximately 1000 human germline sequences and then selecting a suitable template. However, germline sequences provide only a limited view into the overall mutational antibody diversity, which can be addressed by computational humanization, where the animal-derived therapeutic is compared to the distribution of amino acids in human antibody sequences. In one embodiment, computational methods may be employed that compare a query therapeutic sequence to a set of recombined variable region sequences that serve as references in humanization. In one embodiment, computational methods may be employed to assess the "humanness" of a query therapeutic sequence by determining how closely its amino acid content matches the human amino acid distribution.

[0058] In one embodiment, computational parameters may be determined through computational prediction of immune epitopes and ADAs generated for biopharmaceuticals. Generation of immune responses to biopharmaceuticals requires multiple steps beyond reproducing the diversity of human antibody sequences. Even humanized and fully human antibodies may elicit immune responses in patients undergoing such treatment and form ADAs for those patients. The formation of ADAs is a multifactorial problem and may depend, for example, on the patient's genetic background, medical history, protein aggregates in the therapeutic drug, and other degradants. A component of ADA formation is the binding of short peptide fragments derived from the biopharmaceutical to major histocompatibility complex class II (MHC II) molecules. Thus, computational methods may be used to identify potential MHC I and MHC II-bound T cell epitopes, as well as conformational B cell and T cell epitopes.

[0059] In one embodiment, computational parameters related to the biophysical properties of the therapeutic agent may be determined. For example, biophysical properties such as colloidal stability of antibody solutions, concentration-dependent viscosity behavior, and physicochemical degradation. Solubility avoids aggregation, which may lead to loss of activity, antibody degradation, or immunogenicity. From a general perspective, protein aggregation has two aspects: mechanistic and kinetic. The mechanistic aspect focuses on protein instability and identifying potential APRs, mainly hydrophobic patches on the protein surface, which may potentially be nucleation aggregation. Computational methods may be used to predict APRs in biopharmaceuticals according to sequence analysis, for example, the presence of multiple well-defined aggregation-prone motifs, often located within the CDRs. APRs located within these CDRs may contribute to antigen binding. Furthermore, sequence analysis may be used to identify aggregation rate enhancers and mitigating mutations in the protein. Computational methods may be used to predict solubility. Sequence analysis may be used to determine the presence of one or more predictors of solubility and APR in the protein. Computational methods may be used to predict hydrophobicity. The identification of hydrophobic regions may be carried out using homology models.

[0060] At 130, one or more candidate predictive models may be determined. In one embodiment, the experimental parameters and the calculated parameters may be analyzed to determine one or more predictive models that depend on the calculated parameters determined to significantly affect the experimental parameters. One or more computational methods may be used to determine one or more predictive models, including, for example, adaptive context tree weighting, neural networks, CART (classification and regression trees), projection pursuit regression, stepwise regression, linear regression, elastic net, multi-valued models, MARS (multivariate adaptive regression splines), power law, first order graphical LASSO, ridge regression, and general additive models (GAM).

[0061] In one embodiment, stepwise multiple regression (including forward selection, or backward reduction), forced entry, forced removal, and hierarchical multiple regression may be used to determine one or more predictive models. For example, multiple regression analysis may be used to establish a relationship between all independent variables (e.g., experimental parameters) and dependent variables (calculated parameters). The relationship establishes the relative influence of the independent variables. Forward selection (associated with stepwise regression) may then be used to determine the relevance of the independent variables. Forward selection may start with no independent variables in the equation (associated with multiple regression). The independent variables that show the highest correlation or influence with the dependent variable may be added to the equation. The performance of the resulting predictive model may be determined using evaluation techniques. Evaluation techniques (e.g., "goodness of fit" analysis techniques) such as Akaike Information Criterion (AIC), R2, RMS, p-value, F-ratio, standard error, etc. may be used to establish performance characteristics of the relationship. For example, techniques such as R2 that establish the proportion of variance in the dependent variables (e.g., calculated parameters) that is collectively explained by the independent variables (e.g., experimental parameters). Using R2, for example, an assessment can be made as to which relationship best explains the variance of the dependent variable in response to the independent variables. Techniques such as AIC serve as estimators of the within-sample prediction error and therefore the relative quality of the predictive models.

[0062] The forward selection process can be iterative, adding another independent variable (and associated coefficient) to the equation and then evaluating the equation. After all independent variables have been added, the evaluation indices (e.g., AIC, R2) may be compared to determine the equation that best describes the relationship. The variable in the equation that best describes the relationship may be considered to be the most relevant variable, and the other variables may be ignored. For example, a determination may be made as to which variable configuration resulted in the lowest AIC, and / or which variable configuration resulted in the highest R2, or resulted in a significant improvement in R2. In another example, each time an independent variable is added, the relationship may be evaluated to determine whether a significant improvement (e.g., AIC decreased by a significant amount) has been seen. If the evaluation indices have not changed by a significant amount, the process may be stopped, and the independent variables currently forming the relationship may be considered to be the most relevant.

[0063] The backward reduction process (related to stepwise regression) starts with all independent variables in the formula and sequentially removes them, similar to the forward process, to determine the desired relationship. For example, after establishing the relative influence of the independent variables, the least influential independent variables may be removed from the formula. If the resulting AIC does not decrease significantly, the process may be repeated. In one embodiment, stepwise regression may be used to reduce the variables used in constructing the formula or establishing the predictive model.

[0064] At 140, a predictive model may be selected from the candidate predictive models generated in step 130. In one embodiment, a validation technique such as leave-one-out cross-validation (LOOCV) may be used to select a predictive model. LOOCV is a method in which data points are systematically removed from a data set and then their endpoint values ​​are predicted by a relationship derived from the remaining data points (see Cramer et al., Quant. Struct-Act. Relat. 7:18-25, 1998, incorporated herein by reference). Cross-validation is useful to determine the reliability of a relationship, especially when a validation data set is not available. The mean and standard deviation of the error of the predicted LOOCV values ​​from the experimental values ​​may be used as a criterion for comparing and selecting predictive models.

[0065] Once selected, the predictive model can be presented with new computational parameters and make predictions related to the experimental parameters. For example, the predictive model may be trained according to experimental parameters related to mAb solution viscosity and computational parameters related to mAb residue charge values. The predictive model may be presented with the computational parameters of the type on which it was trained, and the predictive model makes predictions related to the experimental parameters on which it was trained.

[0066] For example, the predictive model may be generated according to experimental parameters generated through viscosity measurements consisting of multiple mAb solutions. The viscosity measurements may be obtained through the use of a viscometer. The homology model may be computationally generated and used to determine charge values ​​associated with residues of the mAb. The charge values ​​may be weighted based on whether the residue is determined to be a surface exposed residue. The charge values ​​and / or weighted charge values ​​may be used as computational parameters. The predictive model may be generated according to the experimental and computational parameters. The predictive model may be configured to generate a score indicative of the viscosity. The predictive model may model an untrained mAb (e.g., an input mAb) to generate a charge value / weighted charge value.

[0067] The charge values ​​and / or weighted charge values ​​may be provided to a predictive model that generates a score indicative of viscosity associated with the input mAb. For example, the predictive model may be generated according to experimental parameters generated through aggregation measurements consisting of a plurality of mAb solutions. The aggregation measurements may be obtained through the use of dynamic light scattering. The homology model may be computationally generated and used to determine charge values ​​associated with residues of the mAb. The charge values ​​may be weighted based on whether the residue is determined to be a surface exposed residue. The charge values ​​and / or weighted charge values ​​may be used as computational parameters. The predictive model may be generated according to the experimental parameters and the computational parameters. The predictive model may be configured to generate a score indicative of aggregation. A mAb (e.g., an input mAb) on which the predictive model is not trained may be modeled to generate a charge value / weighted charge value. The charge values ​​and / or weighted charge values ​​may be provided to a predictive model that generates a score indicative of aggregation associated with the input mAb.

[0068] In some embodiments, after providing the optimal predictive model with computationally derived data and determining a viscosity score associated with the query mAb based on the optimal predictive model, appropriate formulation compositions or protein engineering strategies can be adjusted based on the viscosity score to mitigate certain challenges of the drug candidate under development, such as, for example, adjusting the amount of viscosity-lowering agent in the solution associated with the query mAb. In some embodiments, the same can be done for the aggregation score in addition to or instead of the viscosity score. If high viscosity and aggregation scores are calculated for the mAb of interest, different formulation development or protein engineering strategies can be designed. In some embodiments, high aggregation and viscosity scores may indicate the presence of intermolecular interactions that may be determined by a combination of colloidal and conformational interactions. Various excipients that are generally recognized as safe (GRAS) are known to stabilize colloidal and conformational instability, and a combination of such excipients can be utilized to stabilize the mAb structure and reduce viscosity. In some embodiments, high viscosity and low aggregation scores may indicate that the intermolecular interactions are transient and are determined primarily by colloidal interactions. Again, various GRAS excipients are known to reduce electrostatic and hydrophobic interactions between mAbs in solution. In some embodiments, protein engineering can also be used to replace specific amino acids involved in such interactions. In some embodiments, high aggregation scores and low viscosity scores may indicate aggregation primarily due to conformational destabilization. Excipients such as sucrose, various diols and salts have been shown to conformationally stabilize protein structures and can be used in such cases.

[0069] 2, an additional method for generating a predictive model is described. The described method may use machine learning ("ML") techniques to train at least one ML module 230 that is configured to predict a protein viscosity score and / or a protein aggregation score for any antibody based on analysis of one or more training datasets 210 by a training module 220.

[0070] The training data set 210 may include experimental parameters related to direct measurements of antibody solution viscosity and / or antibody aggregation. The experimental parameters are related to computational parameters related to the corresponding antibodies. The computational parameters may be related to charge values ​​of residues on the corresponding antibodies determined via computational modeling. For example, a measurement of the viscosity of a first mAb solution may be related to the charge value of the first mAb. Such data may be derived in whole or in part from experimental data and / or computationally derived data described herein.

[0071] The subset of experimental parameters related to the calculated parameters may be randomly assigned to the training data set 210 or the test data set. In some implementations, the assignment of data to the training data set or the test data set may not be completely random. In this case, one or more criteria may be used during the assignment. In general, any suitable method may be used to assign data to the training data set or the test data set, while ensuring that the yes and no label distributions are somewhat similar in the training data set and the test data set.

[0072] The training module 220 may train the ML module 230 by extracting a feature set from the computational parameters (e.g., labeled with experimental parameters) in the training dataset 210 via one or more feature selection techniques. The training module 220 may train the ML module 230 by extracting a feature set from the training dataset 210 that includes statistically significant features.

[0073] The training module 220 may extract feature sets from the training dataset 210 in a variety of ways. The training module 220 may perform feature extraction multiple times, each time using a different feature extraction technique. In one example, the feature sets generated using different techniques may each be used to generate a different machine learning based classification model 240. For example, the feature set with the highest quality metric may be selected for use in training. The training module 220 may use the feature sets to build one or more machine learning based classification models 240A-240N configured to represent a calculated viscosity and / or a calculated aggregation score for a new mAb (e.g., having unknown viscosity and / or unknown aggregation).

[0074] The training data set 210 may be analyzed to determine any dependencies, associations, and / or correlations between features in the training data set 210 and experimental parameters. The identified correlations may have the form of a list of features. As used herein, the term "feature" may refer to any characteristic of an item of data that can be used to determine whether an item of data is within one or more particular categories. By way of example, the features described herein include Z VL (V L charge relative to the area), Z VH (V H charge relative to the area), Z CL (C L charge relative to the area), Z CH1 (C H charge per area), Z Hinge (charge on the hinge region), Z CH2 (C H 2 charge for the region), Z CH3 (C H 3 charge for the region), Z mAb (total charge), Z * VL (V L Solvent Accessible Area (SAS) adjusted charge for the region), Z * VH (V H SAS adjusted charge for the region), Z *CL (C L SAS adjusted charge for the region), Z * CH1 (C H SAS adjusted charge for one region), Z * Hinge (SAS adjusted charges for the hinge region), Z * CH2 (C H 2 SAS adjusted charge for the region), Z * CH3 (C H SAS adjusted charge for 3 regions), HI (hydrophobicity index), D mAb Or D Fab (total dipole moment), pI Sequence (sequence-based pI), pI Structure (structure-based pI), AP (predicted aggregation propensity), and / or RMSD (root mean square deviation of conformational change).

[0075] The feature selection technique may include one or more feature selection rules, which may include feature generation rules, which may include determining which features occur a threshold number of times in the training data set 210 and identifying those features that meet the threshold as features.

[0076] A single feature selection rule may be applied to select features, or multiple feature selection rules may be applied to select features. Feature selection rules may be applied in a cascading manner, where feature selection rules are applied in a particular order and applied to the results of previous rules. For example, feature generation rules may be applied to the training dataset 210 to generate a first list of features. The final list of features may be analyzed by further feature selection techniques to determine one or more feature sets (e.g., a set of features that can be used to predict viscosity and / or aggregation). Any suitable computational technique may be used to identify the feature sets using any feature selection technique, such as filter methods, wrapper methods, and / or embedding methods. The one or more feature sets may be selected according to a filter method. Filter methods include, for example, Pearson's correlation, linear discriminant analysis, analysis of variance (ANOVA), chi-square, combinations thereof, and the like. The selection of features according to a filter method is independent of any machine learning algorithm. Instead, features may be selected based on scores in various statistical tests for correlation with outcome variables.

[0077] As another example, the one or more feature sets may be selected by a wrapper method. The wrapper method may be configured to use a subset of the features and train the machine learning model using the subset of the features. Features may be added and / or removed from the subset based on inferences drawn from previous models. Wrapper methods include, for example, forward feature selection, backward feature reduction, recursive feature reduction, combinations thereof, and the like. As an example, forward feature selection may be used to identify the one or more feature sets. Forward feature selection is an iterative method that starts with no features in the machine learning model. In each iteration, the feature that best improves the model is added until the addition of a new variable no longer improves the performance of the machine learning model. As an example, backward reduction may be used to identify the one or more feature sets. Backward reduction is an iterative method that starts with all features in the machine learning model. In each iteration, the lowest ranking features are removed until no improvement is observed upon feature removal. Recursive feature reduction may be used to identify the one or more feature sets. Recursive feature reduction is a greedy optimization algorithm that seeks to find the feature subset with the best performance. Recursive feature reduction iteratively builds a model, setting aside the best or worst performing features at each iteration. Recursive feature reduction builds the next model with the remaining features until all features are exhausted. Recursive feature reduction then ranks the features based on their order of reduction.

[0078] As a further example, one or more feature sets may be selected by an embedding method that combines the qualities of filter and wrapper methods. Embedding methods include, for example, least absolute shrinkage and selection operator (LASSO) and ridge regression, which implement a penalty function to reduce overfitting. For example, LASSO regression implements L1 regularization that adds a penalty equivalent to the absolute value of the coefficient magnitude, and ridge regression implements L2 regularization that adds a penalty equivalent to the square of the coefficient magnitude.

[0079] After the feature set is generated by the training module 220, the training module 220 may generate a machine learning based classification model 240 based on the feature set. A machine learning based classification model may refer to a complex mathematical model for data classification that is generated using machine learning techniques. In one example, the machine learning based classification model 240 may include a map of support vectors that represent boundary features. In this example, the boundary features may be selected from and / or represent the highest ranked features in a feature set.

[0080] The training module 220 may use the feature sets determined or extracted from the training dataset 210 to construct the machine learning based classification models 240A-240N. In some examples, the machine learning based classification models 240A-240N may be combined into a single machine learning based classification model 240. Similarly, the ML module 230 may represent a single classifier containing single or multiple machine learning based classification models 240 and / or multiple classifiers containing single or multiple machine learning based classification models 240.

[0081] The features may be combined in a classification model trained using machine learning approaches, such as discriminant analysis; decision trees; nearest neighbor (NN) algorithms (e.g., k-NN models, replicator NN models, etc.); statistical algorithms (e.g., Bayesian networks, etc.); clustering algorithms (e.g., k-means, mean shift, etc.); neural networks (e.g., reservoir networks, artificial neural networks, etc.); support vector machines (SVMs); logistic regression algorithms; linear regression algorithms; Markov models or chains; principal component analysis (PCA) (e.g., for linear models); multi-layer perceptron (MLP) ANN (e.g., for non-linear models); reservoir network replication (e.g., for non-linear models, typically for time series); random forest classification; combinations thereof and / or the like. The resulting ML module 230 may include a decision rule or mapping for each feature to determine antibody viscosity and / or aggregation.

[0082] In one embodiment, the training module 220 may train the machine learning based classification model 240 as a convolutional neural network (CNN), which includes at least one convolutional feature layer and three fully connected layers leading to a final classification layer (softmax), which may be finally applied to combine the outputs of the fully connected layers using a softmax function known in the art.

[0083] The feature(s) and ML module 230 may be used to predict viscosity and / or aggregation from experimental parameters in the test dataset. In one example, the predicted results for each sequence include a confidence level corresponding to the likelihood or probability that the calculated parameters of the mAb in the test dataset are associated with low / high viscosity and / or low / high aggregation. The confidence level may be a value between 0 and 1. In one example, when there are two states (e.g., low and high), the confidence level may correspond to a value p, which refers to the likelihood that the particular mAb belongs to the first state (e.g., low). In this case, the value 1-p may refer to the likelihood that the particular sequence belongs to the second state (e.g., high). Generally, multiple confidence levels may be provided for each mAb in the test dataset, and for each feature if there are three or more states. The best performing feature may be determined by comparing the results obtained for each test mAb to the known experimental parameters for each test mAb. Generally, the best performing feature will have results that closely match the known yes / no promoter states. The best performing features can be used to predict the viscosity and / or aggregation state of the mAb.

[0084] 3 is a flow chart illustrating an example training method 300 for generating the ML model 230 using the training module 220. The training module 220 can implement supervised, unsupervised, and / or semi-supervised (e.g., reinforcement-based) machine learning based classification models 240. The method 300 illustrated in FIG. 3 is an example of a supervised learning method, and variations of this example training method are discussed below, however, other training methods can be implemented similarly to train unsupervised and / or semi-supervised machine learning models.

[0085] The training method 300 may determine (e.g., access, receive, retrieve, etc.) data at step 310. The data may include experimental parameters related to direct measurements of antibody solution viscosity and / or antibody aggregation. The experimental parameters are related to computational parameters related to the corresponding antibodies. The computational parameters may be related to charge values ​​of residues on the corresponding antibodies determined via computational modeling.

[0086] The training method 300 may generate a training data set and a test data set at step 320. The training data set and the test data set may be generated by randomly assigning the computational parameters and the associated experimental parameters to either the training data set or the test data set. In some implementations, the assignment of the computational parameters and the associated experimental parameters as training or test data may not be completely random. As an example, a majority of the computational parameters and the associated experimental parameters may be used to generate the training data set. For example, 75% of the computational parameters and the associated experimental parameters may be used to generate the training data set and 25% may be used to generate the test data set. In another example, 80% of the computational parameters and the associated experimental parameters may be used to generate the training data set and 20% may be used to generate the test data set.

[0087] The training method 300 may, in step 330, determine (e.g., extract, select, etc.) one or more features that can be used by a classifier to distinguish among different classifications of, for example, viscosity and / or aggregation states (e.g., low and high). As an example, the training method 300 may determine the set of features from the calculated parameters and associated experimental parameters. In a further example, the set of features may be determined from data other than the calculated parameters and associated experimental parameters, in either the training data set or the test data set. Such calculated parameters and associated experimental parameters or other data may be used to determine an initial set of features, which may be further reduced using the training data set.

[0088] The training method 300 may train one or more machine learning models at step 340 using one or more features. In one example, the machine learning models may be trained using supervised learning. In another example, other machine learning techniques may be used, including unsupervised learning and semi-supervised. The machine learning models trained at 340 may be selected based on different criteria depending on the problem being solved and / or the data available in the training dataset. For example, machine learning classifiers may be subject to different degrees of bias. Thus, two or more machine learning models may be trained at 340 and optimized, improved, and cross-validated at step 350.

[0089] The training method 300 may select one or more machine learning models to build a predictive model at 360. The predictive model may be evaluated using a test data set. The predictive model may analyze the test data set and generate a predicted viscosity and / or aggregation state at step 370. The predicted viscosity and / or aggregation state may be evaluated at step 380 to determine whether such values ​​achieved a desired level of accuracy. The performance of the predictive model may be evaluated in a number of ways, based on a number of true positive, false positive, true negative, and / or false negative classifications of the multiple data points represented by the predictive model.

[0090] For example, a false positive of a predictive model may refer to the number of times the predictive model erroneously classifies a mAb as low viscosity or low aggregation when it is actually high viscosity or high aggregation. Conversely, a false negative of a predictive model may refer to the number of times the machine learning model classifies a mAb as high viscosity or high aggregation when in fact the mAb is low viscosity or low aggregation. True negatives and true positives may refer to the number of times one or more mAbs are correctly classified by the predictive model. Related to these measurements are the concepts of recall and precision. In general, recall refers to the ratio of true positives to the sum of true positives and false negatives, thereby quantifying the sensitivity of the predictive model. Similarly, precision refers to the ratio of true positives to the sum of true positives and false positives. When such a desired level of precision is reached, the training phase may end and the predictive model (e.g., ML module 230) may be output in step 390; however, when the desired level of precision is not reached, subsequent iterations of the training method 300 may be performed beginning in step 310, with variations such as, for example, taking into account a larger collection of sequence data.

[0091] 4 is a diagram of an exemplary process flow for using a machine learning based classifier to determine whether a mAb is associated with low viscosity and / or low aggregation. As shown in FIG. 4, the unclassified calculated parameters of the mAb 410 may be provided as input to the ML module 230. The ML module 230 may process the unclassified calculated parameters of the mAb 410 using the machine learning based classifier(s) to arrive at a predicted outcome 420.

[0092] The prediction result 420 may identify one or more characteristics of the unclassified calculated parameters of the mAb 410. For example, the classification result 420 may identify the viscosity and / or aggregation state of the unclassified calculated parameters of the mAb 410 (e.g., whether the mAb has low / high viscosity and / or low / high aggregation). EXAMPLES

[0093] A. Computational prediction models of protein solution viscosity and aggregation propensity to facilitate drug development In this study, we developed two predictive models: (1) a solution viscosity prediction model by experimentally measuring the viscosity values ​​of a mixture of 16 IgG1 and IgG4 antibodies and computational full antibody homology modeling of the corresponding antibodies, and (2) an aggregation tendency prediction model by experimentally measuring high molecular weight (HMW) species formation under accelerated thermal stress and computational antigen-binding fragment (Fab) homology modeling and MD simulation of the corresponding Fab region. The approach of this study is to adjust the charge of each residue in the homology model by a weighting factor based on the relative solvent accessible area (SAS) of the exposed residues. With the help of machine learning algorithms, we evaluated the calculated electrostatic and hydrophobic parameters, as well as the conformational changes obtained from the homology model and MD simulation, respectively, to build a robust predictive model for the viscosity and aggregation tendency of protein solutions.

[0094] 1. Method i. Viscosity measurement of protein solutions Sixteen IgG1 and IgG4 antibody solutions, mAb1 to mAb16, formulated at a protein concentration of 150 mg / mL in 10 mM histidine buffer, pH 6.0, were prepared for viscosity measurements. The dynamic viscosity of the solutions was measured using an m-VROC viscometer (Rheosense, San Ramon, CA) at a shear rate of 1420 s -1 and at a rate of 100 μL / min at 20° C. Three viscosity measurements were recorded over a 100 second period.

[0095] ii. Osmotic second virial coefficient (B 22 )measurement Antibody samples were diluted with the corresponding buffer to reach a final protein concentration of 10 mg / mL. Samples were then filtered through a 0.22 μm Millex-GV syringe filter unit (EMD Millipore, Billerica, MA). Light scattering was measured at room temperature using a fully automated compositional gradient multi-angle static light scattering (CG-MALS) instrument with a triple syringe pump Calypso-II sample preparation and delivery unit (Wyatt Technology, Santa Barbara, CA). Both light scattering and protein concentration were measured using a Mini Dawn Treos light scattering instrument (Wyatt Technology, Santa Barbara, CA) equipped with a 658 nm laser and an Optilab Rex refractive index detector (Wyatt Technology, Santa Barbara, CA). Rayleigh ratio light scattering intensity was obtained over a protein concentration range of 2 to 8 mg / mL. Light scattering and protein concentration data were analyzed using Astra 6.1 software (Wyatt Technology, Santa Barbara, CA) to obtain the B 22 To estimate the values, we fit the virial expansion for nonideal solutions, Eq.

[0096]

number

[0097] R θ is the Rayleigh ratio, M w is the molecular weight and c is the protein concentration (mg / mL). 22 denotes the osmotic second virial coefficient that remains unconstrained during data fitting. B 22 provides useful insight into the intermolecular interactions between protein molecules in dilute solutions. 22 A negative value of indicates that the overall interaction between protein molecules is attractive, while a positive value indicates that the overall interaction is repulsive. K in Equation 1 is an optical constant described by Equation 2.

[0098]

number

[0099] n is the refractive index of the solvent (1.33), N A is Avogadro's number (mol -1 ), dn / dc is the increment in refractive index of the protein / solvent pair (0.185 mL / g), and λ is the wavelength of the incident light in vacuum.

[0100] iii. Diffusion interaction parameter (K D )measurement The antibody solution was diluted with 10 mM histidine buffer (pH 6.0) to prepare samples of each mAb at protein concentrations of 10, 5, 2.5, and 0.1 mg / mL. MAb1 was used at 100 mM K due to limited availability of material. D were excluded from the measurements. Samples were centrifuged at 12,000 × g for 5 min and analyzed to remove microbubbles in solution. Light scattering (DLS) was measured using a DynaPro plate reader (Wyatt Technology, Santa Barbara, CA). Fifteen 15-second measurements were collected and averaged to determine the diffusion coefficient for each sample. The interaction parameter K D was calculated based on the following formula:

[0101]

number

[0102] where D is the diffusion coefficient at any protein concentration c, and D0 represents the diffusion coefficient when c is close to 0. Figure 5 shows the K based on fitting lines to the diffusion coefficients at various protein concentrations, using mAb4 as an example. D Show how it was calculated.

[0103] iv. Computational Homology Modeling Full antibody and Fab homology models of 16 IgG1 and IgG4 antibodies were constructed via BIOVIA Discovery Studio 2017 R2. IgG1 and IgG4 antibodies were modeled using the Protein Data Bank (PDB) crystal structures of 1HZH and 5DK3, respectively, as templates and their provided gene sequences. Full antibody models were constructed to calculate physical properties that were later used to develop a predictive model of protein viscosity. Meanwhile, Fab models were used to calculate physical properties and, ultimately, to develop a predictive model of aggregation propensity in MD simulations. The framework for modeling antibody structures is described in detail by Kemmish et al. The energy of the antibody structures was then minimized by geometry optimization using the CHARMM force field for 200 steps. The structures were protonated at pH 6.0, followed by calculation of residue charges and average dipole moments.

[0104] v. Calculated parameters The charge for all regions of the antibody, including the variable and constant regions of both the light and heavy chains (Z VL , Z VH , Z CL , Z CH1 , Z Hinge , Z CH2 , Z CH3 , Z Total , and Z for the Fab model. VL , Z VH , Z CL , Z CH1 , Z Total) were calculated using the procedure described in the Computational Homology Modeling section. The effective charge of each residue was adjusted by considering the relative solvent accessible area (SAS) of that residue within the homology model of the corresponding antibody. In this approach, the charge for each residue was multiplied by a weighting factor calculated using the SAS of the residue relative to the total SAS of either the full antibody or the Fab, depending on the model being used. For example, in the variable light chain, the adjusted charge for each residue was calculated using Equation 3, and the total SAS adjusted charge for this region was calculated using Equation 4. These SAS adjusted charges are calculated using Z for the full model, * VL , Z * VH , Z * CL , Z * CH1 , Z * Hinge , Z * CH2 , Z * CH3 , Z * Total , and Z for the Fab model. * VL , Z * VH , Z * CL , Z * CH1 , Z * Total is labeled as

[0105]

number

[0106] where i = any residue in the variable light chain (VL), and n = the number of residues in the full model or Fab model of the specified antibody.

[0107]

number

[0108] where m = the number of residues in the variable light chain (VL). The hydrophobicity index (HI) of the variable fragment (Fv) was determined using the method described by Sharma et al.

[0109]

number

[0110] where i represents a hydrophobic amino acid, e.g., A, C, F, G, I, L, M, P, V, W, Y; j represents a hydrophilic amino acid, e.g., D, E, H, K, N, Q, R, S, T; n is the number of each amino acid; and E is the Eisenberg scale value of each amino acid. As described above, the average dipole moment of the full model and the Fab model was calculated in the protonated structure at pH 6.0. The isoelectric point (pI) of the antibody used in this study was calculated from both the structural homology model and the sequence. The aggregation propensity (AP) score for each atom is calculated based on the CHARMM force field and SAS patch of exposed hydrophobic residues in a radius of 10 Å. The total aggregation score of each mAb is calculated as the sum of the aggregation scores of all residues in either the full model or the Fab model. A comprehensive list of the computational parameters for the full antibody and Fab models can be found in Table 1, and the values ​​of these computational parameters for each mAb are listed in Figures 6A, 6B for the full antibody model and Figure 7 for the Fab model.

[0111] Table 1. Correlation coefficients (R) obtained from linear regression between the experimentally measured viscosity values ​​and each calculated parameter in the full antibody model, and between the experimentally measured high molecular weight species formation rate per day (%ΔHMW / day) and each calculated parameter in the Fab model. * . JPEG2025032211000012.jpg184170

[0112] i. Mathematical predictive modeling of protein viscosity The experimental viscosity values ​​and the calculated parameters were subjected to a stepwise regression algorithm as dependent and independent variables, respectively. The algorithm generates new linear regression models by adding significant parameters and removing insignificant parameters from the list of parameters, and compares the generated models based on the Akaike Information Criterion (AIC), which is an estimator of the relative quality of a statistical model for any data set based on the number of estimated parameters in the model and the maximum value of the likelihood function of that model. Combinations of parameters in Wilkinson's notation were also considered to generate more models. The results were compared with their R 2 , p-value, and the mean and standard deviation of the error of the predicted viscosity score (PVS) from the experimental values, which are potential predictive models that can be compared together. Furthermore, the models were evaluated by leave-one-out cross-validation (LOOCV) method to ensure the validity of the models in predicting unpredictable data sets. LOOCV excluded one mAb at a time from the training set and evaluated the robustness of the models in predicting the excluded data points. This analysis was repeated 16 times for each model by excluding the calculated and experimental data of different mAbs each time. The mean and standard deviation of the error of the predicted LOOCV viscosity score from the experimental values ​​were used as another criterion for comparing the models. Scripts were developed in the R environment to facilitate the building of predictive models in a high-throughput automated pipeline.

[0113] ii. Accelerated thermal stress stability and aggregation kinetics measurements Accelerated stability studies are routinely performed during formulation development in pharmaceutical companies. In this study, the effect of thermal stress on the overall stability was evaluated by incubating 14 mAb samples used in the viscosity study at 40°C and 75% relative humidity for 0, 7, 14, and 28 days. Due to limited material availability for mAb9 and mAb16, these two candidates were excluded from the current data set for the development of a predictive model for aggregation propensity. Size exclusion chromatography (SEC) was used to measure the amount of high molecular weight (HMW) species formation. The relative percentage of HMW formation at 7, 14, and 28 days, %ΔHMW, was calculated by comparing it to day 0. Additionally, the rate of %ΔHMW formation per day was calculated based on the %ΔHMW divided by 28 for the 28-day data point.

[0114] iii. Molecular Dynamics (MD) Simulation To include conformational changes in the Fab region in the prediction model of aggregation propensity, MD simulations of the Fab models were utilized. The atoms of the minimized Fab structures of the 14 antibodies were assigned to CHARMM36 force field parameters by structurally matching each residue to its template. These structures were then explicitly solvated within a truncated octahedral box of TIP3P water molecules. The counterion Na + and Cl - was added to the explicitly solvated systems to neutralize the systems with an ion concentration of 0.145 mol / L. For each simulation, the energy of the system was first minimized using a steepest descent algorithm for 1000 steps, followed by an additional 2000 steps of minimization using ABNR to remove large distortions in the system. The systems were gradually heated from 50.0 to 300.0 K in 4 ps with an interval of 50 K under a constant volume (NVT) ensemble with a time step of 2.0 fs. Each system was then further equilibrated for 10 ps with a target temperature of 300.0 K and a time step of 2.0 fs under an isotropic pressure of 1.0 bar. Finally, each system was simulated for 2000 ps (i.e., 2.0 ns) with a time step of 2.0 fs at a constant temperature of 300.0 K and a pressure of 1.0 bar.

[0115] Long-range electrostatics were used using the Particle Mesh Ewald (PME) method, with a 10 Å cutoff distance for van der Waals interactions. The SHAKE algorithm was applied to each simulation, constraining bond lengths for all hydrogen atoms to allow a time step of 2.0 fs. Three simulations of 2.0 ns each (6.0 ns overall) were performed for each system, differing only in the initial distribution of velocities, to allow scrutiny of the reproducibility of the results. Trajectories, time-varying atomic coordinates, for each simulation were captured every 1.0 ps (i.e., 2000 conformations overall for each simulation). The backbone root mean square deviation (RMSD) of the conformational structure relative to the initial structure after rigid-body alignment in each simulation was calculated as a descriptor of conformational stability.

[0116] iv. Mathematical predictive modelling of agglomeration tendencies As previously mentioned, protein aggregation is a collective effect of colloidal stability (i.e., intermolecular interactions) and conformational stability (i.e., changes in protein structure). To build a predictive model of aggregation propensity, the experimental rate of HMW formation (i.e., %ΔHMW / day), the physically calculated parameters of the Fab model (Table 1) as a descriptor of colloidal stability, and the averaged RMSD in MD simulations as a descriptor of conformational stability were used in the same protocol as described in the viscosity section. In summary, a stepwise regression algorithm was used to create the most statistically significant models correlating the calculated colloidal and conformational parameters with the measured aggregation kinetics. These models were evaluated using AIC numbers, p-values, and R 2 , adjusted R 2 ,The mean and standard deviation of the absolute error, the performance of LOOCV, and the,structural symmetry of the antibody structures were compared with each other.

[0117] 2.Results i. Viscosity of the protein solution Viscosity values ​​were measured for 16 mAbs at a protein concentration of 150 mg / mL. Overall, the viscosity values ​​show a broad distribution ranging from 5.5 to 32.0 cP (Table 2 and Figure 8). According to the dataset, IgG1 antibodies tend to show lower viscosity values ​​compared to IgG4 candidates (Figure 8).

[0118] Table 2. Measured viscosity and calculated predicted viscosity score (PVS) values ​​for the 16 mAbs used in this study. Viscosity values ​​were measured in formulations with a protein concentration of 150 mg / mL, 10 mM histidine buffer, pH 6.0. PVS values ​​were calculated based on Equation 5. The absolute error between the PVS and measured viscosity values ​​was calculated. The absolute error for PVS and leave-one-out cross-validation (LOOCV) methods is also shown. JPEG2025032211000013.jpg140170

[0119] ii.B22 and KD show a strong correlation with each other, but no correlation with viscosity Osmotic second virial coefficient (B 22 ) and the diffusion interaction parameter (K D ) were measured for 16 and 15 mAbs, respectively. For the current dataset, B 22 The value is -1.461 x 10 -05 ~2.939×10 -04 mol ml g -2 It varies between K D The values ​​vary between -11.604 and 61.114 mL / g (Figures 9A-9C). Based on the data set, IgG1 antibodies showed higher positive B 22 Value and K D There is a tendency for the value to show B 22 value and K D The values ​​are strongly correlated with each other for the current data set, with a linear correlation coefficient (R) of 0.99 (Figures 9A-9C), an observation consistent with previously published studies by others in the field.

[0120] B 22 and K. DBoth measurements are measures of pairwise interactions, which predominate primarily at dilute concentrations. However, as concentration increases, higher order interactions involving multiple molecules also contribute significantly to solution viscosity. Thus, the B measured in dilute solutions 22 value and K D The B22 value is not a direct measure of protein-protein interactions at high concentrations. However, there is debate in the literature regarding the validity of using B22 and Kd values ​​at dilute concentrations as predictors of viscosity values ​​at high protein concentrations. To assess this, the B 22 or K D A linear correlation between either B or the measured viscosity values ​​was determined (Figures 10A-10B). Based on the data set, B 22 or K D There is a directional decreasing tendency (i.e., a negative correlation) between either of the values ​​and the viscosity value, but there is no strong correlation (Figures 10A-10B). 22 and K. D is a poor predictor of protein solution viscosity as mAb concentration increases.

[0121] iii. Each single selected calculation parameter contributes to the prediction of the overall viscosity value The linear correlation coefficient (R) values ​​obtained from the regression lines between the measured viscosity values ​​and the calculated parameters from the full antibody homology model of the 16 mAbs are shown in Table 1. Additionally, Figures 11A-11C show plots of the linear relationship between the experimental viscosity values ​​and the calculated parameters. The correlation R values ​​vary between -0.68 and 0.54. The calculated parameters (i.e., Z * VL , Z * CL , Z * Hinge , Z * CH2 , Z * CH3The parameters (H, H I) were chosen based on the stepwise protocol described in the mathematical modeling section and the structural symmetry of the antibody. Although each of these parameters contributes to the prediction of the overall viscosity value, no single calculated parameter can predict the viscosity value alone, as observed for the moderate R values. This was indeed expected, since the nature of viscosity is a multivariate phenomenon with intermolecular interactions defined by the hydrophobic and electrostatic properties of different regions.

[0122] iv. Prediction Model of Protein Viscosity: Predicted Viscosity Score (PVS) A final predictive model of protein viscosity was selected based on the stepwise protocol and selected calculated parameters as described in the previous section. The predicted viscosity score (PVS) is the V of the full antibody model. L area, C L Area, hinge area, C H 2 Area and C H This is a prediction model for protein viscosity that takes into account the solvent accessible surface area (SAS) for three regions, the adjusted charge, and the hydrophobicity of the variable region (Equation 5).

[0123] Table 3 shows constants C0 to C6 in the PVS model.

[0124]

number

[0125] Table 3: Constant coefficients for Equation 5, predicted viscosity score (PVS). JPEG2025032211000015.jpg65170

[0126] The PVS model is 2 is 0.884, adjusted R 2is 0.807. The correlation R value of 0.94 indicates a strong correlation between PVS and measured viscosity values ​​in the current data set, and the adjusted R value of 0.90 indicates that the model was fitted to any data set by considering the number of parameters in the model without concerns of overfitting the data. The p-value of the PVS model is 0.0009, indicating strong evidence against the null hypothesis that the PVS model cannot predict viscosity values ​​with 95% confidence (p-value < 0.05). The mean absolute error of 2.7 as well as the standard deviation of 1.8 observed between PVS and experimental values ​​attest to the validity of the model. The minimum and maximum residuals between PVS and measured viscosity are -7.9 and 3.9, respectively (Table 2). The linear regression line between the calculated PVS values ​​and the measured viscosity values ​​indicates that the majority of the data points are within the 95% confidence interval (Figure 12). More importantly, the PVS model performs well in the LOOCV analysis, with a mean absolute error of 4.6 and standard deviation of 2.3 between the PVS based on the LOOCV method and the experimental viscosity values ​​(Table 2). 2 The values ​​ranged from 0.863 to 0.925, and the adjusted R 2 The values ​​ranged from 0.760 to 0.868. These results confirm that the PVS represents a statistically significant predictive model between the selected computational parameters and the experimental viscosity values.

[0127] v. Thermal stress stability results The 14 mAbs were incubated at 40 °C and 75% relative humidity. Figure 13 shows a representative SEC signal for mAb3 over 0 and 28 day incubation periods, as well as the increase in the HMW peak as a result of aggregate formation. The relative percentage of HMW formation, %ΔHMW, at 7, 14, and 28 days compared to 0 days is shown in Figure 14 for the 14 mAbs used in this study. The longer the mAb samples are incubated, the more aggregates form (Figure 14). The rate of formation of %ΔHMW per day, calculated based on the %ΔHMW divided by 28 for the 28 day data point, ranges from 0.0564 to 0.1600 (Table 4 and Figure 15). IgG1 antibody molecules, which tend to have lower viscosity values, tend to have a higher %ΔHMW / day on average compared to IgG4 candidates in the current data set (Figure 15).

[0128] Table 4. Measured rates of high molecular weight species formation per day (%ΔHMW / day) and calculated predicted aggregation score (PAS) values ​​for the 14 mAbs used in this study. Due to limited material availability for mAb9 and mAb16, these two candidates were excluded from the current data set for the development of a predictive model for aggregation propensity. The %ΔHMW / day values ​​were calculated after 28 days of incubation at 40° C. and 75% relative humidity. The PAS values ​​were calculated based on Equation 6. The absolute errors between the PAS values ​​and the measured %ΔHMW / day values ​​were calculated. The absolute errors for the PAS and leave-one-out cross-validation (LOOCV) methods are also shown. JPEG2025032211000016.jpg121170

[0129] vi. Further verification of PVS To further test the predictive potential of the PVS model (Equation 5), four IgG1 and IgG4 mAbs (Table 5) that were not part of the 16 mAbs used to develop the predictive model were evaluated. The viscosities of these four mAbs were measured in the same formulations using the same protocol as previously described. The measured viscosity values ​​range from 4.1 to 22.0 cP (Table 5). The structures of these four mAbs were modeled following the protocol described in this work. Hydrophobic and electrostatic parameters were calculated for utilization in the PVS model (Table 5). The absolute errors between the PVS and experimental viscosity values ​​ranged from 2.8 to 5.6 (Table 5), indicating that the PVS model can accurately predict the viscosity values ​​of mAbs not included in the training dataset.

[0130] Table 5: Calculation parameters, measured viscosity values, and predicted viscosity scores (PVS) of 4 mAbs that were not part of the 16 mAbs were used to develop the viscosity prediction model. * VL , Z * CL , Z * Hinge , Z * CH2 and Z * CH3 are V L area, C L Area, hinge area, C H where HI is the adjusted charge of the solvent accessible area for the two regions, and HI is the hydrophobicity index. The PVS values ​​were calculated based on Equation 5, and the viscosity values ​​were measured for formulations with a protein concentration of 150 mg / mL, 10 mM histidine buffer, pH 6.0. The absolute error between the PVS values ​​and the measured viscosity values ​​was calculated. JPEG2025032211000017.jpg61170

[0131] vii.MD Simulation Results For each Fab model of the 14 mAbs, three separate MD simulations were performed to assess the consistency of the observations. The RMSD of the conformational structure relative to the initial structure of each mAb was plotted over a simulation time of 2.0 ns (Figures 16A-16C), showing that the overall simulations for each mAb were reproducible. For each mAb, the RMSD values ​​at each time point were averaged over the three simulations, and the averaged RMSD was plotted against the simulation time of 2.0 ns in Figure 17. Furthermore, Figures 18A-18C show the averaged RMSD of each mAb in separate plots. The averaged RMSD values ​​over the three simulations of each mAb were averaged over the last 1.5 ns to obtain a single number for the averaged RMSD of each mAb. For the Fab regions of the 14 mAbs used in this study, the averaged RMSD ranges from 1.785 to 3.159 Å (Figure 7).

[0132] viii. Each single selected calculation parameter contributes to the prediction of the overall aggregation tendency The calculated parameters from the Fab homology model and the RMSD values ​​from the MD simulations are the descriptors of colloidal and conformational stability, respectively. The linear correlation coefficient (R) values ​​obtained from the regression lines between the measured %ΔHMW / day values ​​and the calculated parameters or RMSD values ​​obtained from the MD simulations of the 14 mAbs are shown in Table 1. Furthermore, Figures 19A-19C show plots of the linear relationship between the experimental %ΔHMW / day values ​​and the calculated parameters including RMSD. The correlation R values ​​vary between -0.59 and 0.72. The final prediction model (i.e., Z * VL , Z * CH1 , RMSD, H.I., D Fab , and P.I. SequenceThe calculated parameters used in () were selected based on the stepwise protocol described in the mathematical modeling section and the structural symmetry of the Fab region. Although each of these parameters contributes to the prediction of the overall %ΔHMW / day value, no single calculated parameter can independently predict the %ΔHMW / day value, as observed by the moderate R values. This was expected as a result of the nature of aggregation being a multivariate phenomenon involving both colloidal and conformational stability terms.

[0133] ix. Prediction model of aggregation tendency: Predicted Aggregation Score (PAS) A final predictive model of aggregation propensity was selected based on the stepwise protocol and selected calculated parameters as described in the Test Methods section. The Predicted Aggregation Score (PAS) is a predictive model of protein aggregation kinetics that includes both colloidal and conformational calculated descriptors of the Fab region. This model is a predictive model of the V L Area and C H1 The SAS adjusted charge for the region, the averaged backbone RMSD of the conformational changes relative to the initial structure, the hydrophobicity of the variable region, the dipole moment of the Fab region, and the isoelectric point of the antibody obtained from the sequence were taken into consideration (Equation 6). The constants C0 to C6 in the PAS model are shown in Table 6.

[0134]

number

[0135] Table 6: Constant coefficients for Equation 6, predicted cohesion score (PAS). JPEG2025032211000019.jpg70170

[0136] The PAS model is R 2 is 0.883, adjusted R 2is 0.782. The correlation R value of 0.94 indicates a strong correlation between PAS and measured %ΔHMW / day values ​​in the current data set, and the adjusted R value of 0.88 indicates that the model fitted well to any data set by considering the number of parameters in the model without concerns of overfitting the data. The p-value of the PAS model is 0.0057, indicating strong evidence against the null hypothesis that the PAS model cannot predict %ΔHMW / day values ​​with 95% confidence (p-value < 0.05). The mean absolute error of 0.0084 and standard deviation of 0.0083 observed between PAS and experimental values ​​attest to the validity of the model. The minimum and maximum residual errors between PAS and measured %ΔHMW / day are -0.0290 and 0.0216, respectively (Table 3).

[0137] The linear regression line between the calculated PAS values ​​and the measured %ΔHMW / day values ​​shows that the majority of the data points are within the 95% confidence interval (Figure 20). More importantly, the PAS model performs well in the LOOCV analysis, with a mean absolute error of 0.0173 and standard deviation of 0.0151 between the PVS based on the LOOCV method and the experimental %ΔHMW / day (Table 4). During LOOCV, R 2 The values ​​ranged from 0.858 to 0.949, with adjusted R 2 The values ​​ranged from 0.716 to 0.897. These results confirm that PAS represents a statistically significant predictive model between the calculated parameters of selected colloidal structures and conformations and the experimental %ΔHMW values.

[0138] 3. Discussion Monoclonal antibody production is increasing in the pipelines of biopharmaceutical companies. The trend from IV to SC administration poses challenges in formulation development of these candidates, which require administration of higher potencies due to viscosity and mAb aggregation issues. More robust methods should be developed to enable early stage prediction of viscosity and aggregation propensity in formulation development and drug discovery. The combination of computational tools and experimental methods shows promise in solving these challenges. Here, a scheme is developed that utilizes the power of homology modeling and MD simulation to generate predictive models of viscosity and aggregation propensity of antibody solutions. These models, PVS and PAS, allow for the comparison of different antibodies together, without the need for physical materials, in drug development and even in the early stages of drug discovery.

[0139] Compared with similar predictive models in the literature for the viscosity and aggregation tendency of protein solutions, the PVS and PAS models developed in this study showed lower prediction errors and higher R 2 Value, adjusted R 2 The values, p-values, and absolute error values ​​indicate that the PVS and PAS models are more reliable based on the LOOCV analysis. The robustness of the PVS and PAS models is obtained by utilizing novel computational parameters that adjust the charge distribution for the SAS of each residue in different regions of the antibody and by taking into account the symmetry of its structure. In the PVS model, the computed parameters of electrostatics and hydrophobicity from the antibody homology model are considered to reflect the intra- and intermolecular interactions in the protein solution.

[0140] This is the first time that both colloidal and conformational stability parameters have been used in a prediction model of aggregation tendency via all-atom MD simulations. Thus, the PAS model can predict aggregation tendency more realistically by considering stability in atomic details. It goes without saying that the constant coefficients of the PVS and PAS prediction models developed for viscosity and aggregation tendency are specific to the buffer system and respective protein concentration used in this study. However, the overall scheme and calculated parameters can be extended to other buffer systems and protein concentrations.

[0141] The robustness and accuracy of predictive models in machine learning algorithms and statistical methods depend on the number of data points in the training and validation datasets. However, the nature of experimental viscosity, aggregation measurements, and limited availability of physical materials limit the ability of researchers in this field to obtain a large number of data points. Therefore, all data points were used as training datasets and LOOCV analysis was performed to evaluate the predictive potential of the models. The PVS and PAS models were developed based on the viscosity and aggregation measurements of 16 and 14 mAbs, respectively. More data points measured under the same conditions can be added to the datasets developed for these models to improve the robustness and accuracy of these predictive models. The new data points can also be used as validation datasets to evaluate the further predictive potential of the PVS and PAS models. In addition, a stepwise regression algorithm was utilized in this study to generate predictive models from the calculated parameters. Further statistical and machine learning algorithm techniques such as LASSO (least absolute shrinkage and selection operator) regression and random forest regression can be explored to develop more robust models.

[0142] As mentioned before, MD simulations were performed on the Fab region of an antibody to consider the conformational stability in the aggregation-prone model. The protocol described in this work can be extended to perform MD simulations on a full antibody to generate calculated parameters of the conformation for the full antibody model. These parameters may improve the validity and reliability of the aggregation-prone model. In addition, full antibody all-atom and coarse-grained MD simulations can provide more hints to the intra- and intermolecular interactions of antibody molecules. As an example, Cloutier et al. analyzed the effects of excipients on aggregation and viscosity through all-atom MD simulations for three IgG1 mAbs. Kastelic et al. performed coarse-grained MD simulations to evaluate the binding interactions of fragment antigen (Fab-Fab) or crystallizable fragment (Fab-Fc) and proposed a strategy to control the viscosity of antibody solutions through the control of their binding sites.

[0143] In the current study, each mAb was dissolved in a truncated octahedral water box to minimize the number of water atoms in favor of heavy computational simulations. Even with this scheme, each solvated full antibody or Fab model consists of approximately 275,000 or 51,000 atoms, respectively. Such large systems require a large amount of computational power, and simulation time may be limited by the available infrastructure. Advances in graphics processing units (GPUs) allow longer MD simulations for Fab regions and full antibodies. Longer MD simulations may provide more hints to the conformation and inherent instability of antibodies in solution, leading to a better understanding of intra- and intermolecular interactions. In addition, other schemes, such as rectangular water boxes, can be utilized to account for the interactions of more water molecules with a given antibody. The box shape used in the simulations may affect the dynamic behavior of the protein and computational properties.

[0144] Furthermore, all MD simulations in this study were performed in aqueous solutions. This is the most common approach in MD simulations of biologics, as most force fields are optimized for water interactions. However, performing MD simulations in a buffer environment similar to the final drug product can yield computational parameters that are more relevant to experimental viscosity and stability measurements. When ranking antibodies, as long as all mAb candidates are treated in the same buffer system, they can be used in predictive models and compared to each other.

[0145] B. Model Validation Validation experiments were performed for both PVS (predicted viscosity score) and PAS (predicted aggregation score) models and algorithms using a set of 10 mAbs. Statistical correlation of data between the experimental data set and the predicted scores was used to validate the viscosity and aggregation prediction algorithms.

[0146] 1. Test Method The prediction model was validated using experimental data for dynamic viscosity (cP) at 40°C and total aggregation rate for 10 additional mAbs. The experimental data was blinded to the users to remove known and unknown biases. The predicted scores for viscosity and aggregation were then compared to the experimental data and correlated using a linear regression model. Given the small size of the dataset, a correlation score above 0.75 is considered an acceptable correlation.

[0147] Model validation using a test set (data-blinded correlation) for dynamic viscosity is shown in Figure 21. The PVS (Equation 5) and PAS (Equation 6) models were validated using data from 10 mAbs (a mix of IgG1 and IgG4). Data was blinded from the users to ensure unbiased validation. The color coding for risk ranking (separated by dashed lines and labeled for color) is based on past development goals across the biopharmaceutical industry and does not reflect any regulatory requirements. JPEG2025032211000020.jpg71170

[0148] Model validation using a test set (data blinded correlation) for high molecular weight (HMW) species formation is shown in Figure 22. %ΔHMW was calculated as follows:

[0149]

number

[0150] The PVS and PAS models were validated using data from 10 mAbs (a mix of IgG1 and IgG4). Data were blinded to users to ensure unbiased validation.

[0151] The color coding for risk ranking (separated by dashed lines and labeled with color) is based on historical development goals across the biopharmaceutical industry and does not reflect any regulatory requirements. JPEG2025032211000022.jpg67170

[0152] 2.Results The prediction scores for both flocculation and viscosity were highly correlated to the validation experimental data (R2 values ​​>0.8). Strong statistical correlations further improve confidence in both the predictive models and their underlying AI algorithms.

[0153] FIG. 23 is a block diagram depicting an environment 2300 including a non-limiting example of a computing device 2301 and a server 2302 connected through a network 2304. In one embodiment, some or all steps of any described method can be performed on a computing device described herein. The computing device 2301 can include one or more computers configured to store one or more of the experimental data 2320, the computationally derived data 2322, the prediction module 2326 (e.g., the ML module 230, including any supplemental training modules), and the like. The server 2302 can include one or more computers configured to store the experimental data 2320 and / or the computationally derived data 2322. The multiple servers 2302 can communicate with the computing device 2301 through the network 2304. In one embodiment, the server 2302 can include a repository for data generated by one or more experiments.

[0154] In terms of hardware architecture, the computing device 2301 and the server 2302 may be digital computers that generally include a processor 2308, a memory system 2310, an input / output (I / O) interface 2312, and a network interface 2314. These components (2308, 2310, 2312, and 2314) are communicatively coupled via a local interface 2316. The local interface 2316 may be, for example, but not limited to, one or more buses or other wired or wireless connections known in the art. The local interface 2316 may have additional elements (omitted for simplicity) to enable communication, such as controllers, buffers (caches), drivers, repeaters, and receivers. Additionally, the local interface may include address, control, and / or data connections to enable appropriate communication between the aforementioned components.

[0155] The processor 2308 may be a hardware device for executing software, particularly stored in the memory system 2310. The processor 2308 may be any custom-made or commercially available processor, a central processing unit (CPU), a coprocessor among several processors associated with the computing device 2301 and the server 2302, a semiconductor-based microprocessor (in the form of a microchip or chipset), or generally any device for executing software instructions. When the computing device 2301 and / or the server 2302 are in operation, the processor 2308 may be configured to execute software stored in the memory system 2310 to communicate data to and from the memory system 2310, and generally control the operation of the computing device 2301 and the server 2302 in accordance with the software.

[0156] The I / O interface 2312 can be used to receive user input from and / or provide system output to one or more devices or components. User input may be provided, for example, via a keyboard and / or a mouse. System output may be provided via a display device and a printer (not shown). The I / O interface 2312 may include, for example, a serial port, a parallel port, a small computer system interface (SCSI), an infrared (IR) interface, a radio frequency (RF) interface, and / or a universal serial bus (USB) interface.

[0157] The network interface 2314 can be used to transmit and receive from the computing device 2301 and / or the server 2302 over the network 2304. The network interface 2314 may include, for example, a 10BaseT Ethernet adapter, a 100BaseT Ethernet adapter, a LAN PHY Ethernet adapter, a Token Ring adapter, a wireless network adapter (e.g., WiFi, cellular, satellite), or any other suitable network interface device. The network interface 2314 may include address, control, and / or data connections to enable appropriate communication over the network 2304.

[0158] The memory system 2310 may include any one or combination of volatile memory elements (e.g., random access memory (RAM, such as DRAM, SRAM, SDRAM, etc.)) and non-volatile memory elements (e.g., ROM, hard drives, tape, CD-ROM, DVD-ROM, etc.). Additionally, the memory system 2310 may incorporate electronic, magnetic, optical, and / or other types of storage media. It should be noted that the memory system 2310 may have a distributed architecture where various components are located remotely from each other but can be accessed by the processor 2308.

[0159] The software in the memory system 2310 may include one or more software programs, each of which includes an ordered list of executable instructions for implementing logical functions. In the example of FIG. 23, the software in the memory system 2310 of the computing device 2301 may include experimental data 2320, computationally derived data 2322, a prediction module 2326, and a suitable operating system (O / S) 2318. In the example of FIG. 23, the software in the memory system 2310 of the server 2302 may include experimental data 2320, computationally derived data 2322, and a suitable operating system (O / S) 2318. The operating system 2318 essentially controls the execution of other computer programs and provides scheduling, input-output control, file and data management, memory management, and communication control, and related services.

[0160] For purposes of illustration, application programs and other executable program components, such as the operating system 2318, are illustrated herein as separate blocks, with the understanding that such programs and components may reside at various times in different storage components of the computing device 2301 and / or the server 2302. An implementation of the prediction module 2326 may be stored on or transmitted through some form of computer-readable medium. Any of the methods of the present disclosure may be performed by computer-readable instructions embodied on a computer-readable medium. A computer-readable medium may be any available medium that can be accessed by a computer. By way of example, and not intended to be limiting, computer-readable media may include "computer storage media" and "communications media." A "computer storage medium" may include volatile and non-volatile removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Exemplary computer storage media may include RAM, ROM, EEPROM, flash memory or other storage technology, CD-ROM, digital versatile disks (DVD) or other optical storage devices, magnetic cassettes, magnetic tapes, magnetic disk storage devices or other magnetic storage devices, or any other medium that can be used to store the desired information and that can be accessed by a computer.

[0161] In one embodiment, the prediction module 2326 may be configured to perform the method 2400 shown in FIG. 24. The method 2400 may be performed in whole or in part by a single computing device, multiple electronic devices, and the like. The method 2400 may include, at 2410, determining experimental data associated with one or more monoclonal antibodies (mAbs). The one or more mAbs may include one or more of an IgG1 antibody or an IgG4 antibody. The experimental data may include experimental viscosity data. The experimental viscosity data may include one or more of a dynamic viscosity value or a kinematic viscosity value.

[0162] Determining the experimental data relating to the one or more mAbs may include measuring at least one of a dynamic viscosity value or a kinetic viscosity value based on a solution of each of the one or more mAbs and a viscometer.

[0163] The experimental data can include experimental aggregation data. The experimental aggregation data can include high molecular weight (HMW) speciation data for each mAb of the one or more mAbs. Determining the experimental data relating to the one or more mAbs can include measuring the amount of HMW speciation over time based on solution and size exclusion chromatography (SEC) of each of the one or more mAbs.

[0164] At 2420, computationally derived data associated with the one or more mAbs is determined, the computationally derived data including one or more computational parameters weighted based on the surface area exposed (ASA) of one or more residues of the one or more mAbs. The computationally derived data may include charge data associated with one or more regions associated with the sequences of the one or more mAbs, modified charge data associated with one or more regions based on the solvent accessible area of ​​residues in a homology model of the one or more mAbs, hydrophobicity index (HI), dipole moment, or isoelectric point (pI). Determining computationally derived data associated with the one or more mAbs may include full antibody homology modeling of the sequences of the one or more mAbs, or antigen binding fragment (Fab) region modeling of the Fab sequences of the one or more mAbs.

[0165] Determining computationally derived data associated with the one or more mAbs may include determining one or more charge values ​​associated with one or more residues in one or more regions of the one or more mAbs based on a homology model of the one or more mAbs; determining a solvent accessible surface (SAS) of one or more residues in the one or more regions based on the homology model of the one or more mAbs; adjusting one or more charge values ​​associated with the one or more residues based on a weighting factor calculated using the SAS of the one or more residues relative to a total SAS associated with the one or more mAbs; and determining a charge value associated with each region of the one or more regions based on the homology model of the one or more mAbs and the adjusted one or more charge values ​​associated with the one or more residues.

[0166] The computationally derived data may include charge data associated with one or more regions associated with the sequence of the one or more mAbs, modified charge data associated with one or more regions based on the solvent accessible area of ​​residues in a homology model of the one or more mAbs, hydrophobicity index (HI), dipole moment, isoelectric point (pI), aggregation tendency (AP), or a conformational stability descriptor. The conformational stability descriptor may include a backbone root mean square deviation (RMSD) of the conformation relative to an initial structure after rigid body alignment. Determining the computationally derived data associated with the one or more mAbs may include one or more molecular dynamics (MD) simulations associated with the one or more mAbs.

[0167] At 2430, a plurality of candidate predictive models are determined based on the experimental data and the computationally derived data. Determining the plurality of candidate predictive models based on the experimental data and the computationally derived data may include identifying one or more experimental parameters of the experimental data as dependent variables, identifying one or more computational parameters of the computationally derived data as independent variables, and determining the plurality of candidate predictive models based on the dependent variables and based on the independent variables based on a stepwise regression algorithm.

[0168] At 2440, an optimal predictive model is determined from the multiple candidate predictive models. Determining an optimal predictive model from the multiple candidate predictive models may include determining an Akaike Information Criterion (AIC) score for each candidate predictive model of the multiple candidate predictive models, and determining the candidate predictive model of the multiple candidate predictive models associated with the highest AIC score as the optimal predictive model.

[0169] Determining an optimal predictive model from the plurality of candidate predictive models may include determining as the optimal predictive model the candidate predictive model of the plurality of candidate predictive models associated with the smallest error in predicting the viscosity score of the excluded mAb from the experimental data and the computationally derived data.

[0170] Determining an optimal predictive model from the plurality of candidate predictive models may include determining as the optimal predictive model the candidate predictive model of the plurality of candidate predictive models associated with the smallest error in predicting the aggregation score of the excluded mAb from the experimental data and the computationally derived data.

[0171] In 2450, the optimal predictive model is output. Method 2400 may also include receiving computationally derived data associated with the query mAb, providing the computationally derived data to a best fit predictive model, and determining a viscosity score associated with the query mAb based on the best fit predictive model. Method 2400 may include adjusting an appropriate formulation composition or protein engineering strategy to mitigate a particular challenge of the drug candidate under development, such as, for example, adjusting an amount of a viscosity-lowering agent in a solution associated with the query mAb, based on the viscosity score.

[0172] The method 2400 may also include receiving computationally derived data related to the query mAb, providing the computationally derived data to a best-fit predictive model, and determining an aggregation score based on the best-fit predictive model.

[0173] In one embodiment, the prediction module 2326 may be configured to perform the method 2500 shown in FIG. 25. The method 2500 may be performed in whole or in part by a single computing device, multiple electronic devices, and the like. The method 2500 may include, at 2510, receiving computationally derived data related to the monoclonal antibody (mAb). The computationally derived data may include computationally derived viscosity data. The computationally derived viscosity data may include one or more of a dynamic viscosity value or a kinematic viscosity value.

[0174] At 2520, the computationally derived data is provided to a predictive model. At 2530, a viscosity score associated with the mAb is determined based on the predictive model. Method 2500 may also include adjusting an appropriate formulation composition or protein engineering strategy to alleviate a particular challenge of a drug candidate under development, such as, for example, adjusting the amount of a viscosity-lowering agent in the solution associated with the query mAb, based on the viscosity score.

[0175] Method 2500 may also include receiving sequence data associated with the mAb and determining computationally derived data based on the sequence data. Method 2500 may also include receiving computationally derived data associated with the query mAb, providing the computationally derived data to a best fit predictive model, and determining a viscosity score associated with the query mAb based on the best fit predictive model.

[0176] In one embodiment, the prediction module 2326 may be configured to perform the method 2600 shown in FIG. 26. The method 2600 may be performed in whole or in part by a single computing device, multiple electronic devices, and the like. The method 2600 may include, at 2610, receiving computationally derived data related to a monoclonal antibody (mAb). The computationally derived data may include computationally derived aggregation data. The computationally derived aggregation data may include high molecular weight (HMW) speciation data for the mAb. The computationally derived data may include charge data related to one or more regions related to the sequence of the mAb, modified charge data related to one or more regions based on the solvent accessible area of ​​residues in a homology model of the mAb, hydrophobicity index (HI), dipole moment, isoelectric point (pI), aggregation propensity (AP), or conformational stability descriptors.

[0177] At 2620, the computationally derived data is provided to a predictive model. At 2630, an aggregation score associated with the mAb is determined based on the predictive model. Method 2600 may also include receiving sequence data associated with the mAb and determining computationally derived data based on the sequence data.

[0178] Method 2600 may also include determining an optimal predictive model from the plurality of candidate predictive models associated with a minimum error in predicting an aggregation score associated with the mAb. Method 2600 may also include receiving computationally derived data associated with the query mAb, providing the computationally derived data associated with the query mAb to a best fit predictive model, and determining an aggregation score associated with the query mAb based on the best fit predictive model.

[0179] In light of the described apparatus, systems, and methods and variations thereof, the present specification now describes certain more particularly described embodiments of the present invention below, but these specifically recited embodiments should not be construed as having any limiting effect on any different claims that incorporate different or more general teachings set forth herein, nor should the "specific" embodiments be construed as being limited in any way other than in the inherent meaning of the language literally used therein.

[0180] Embodiment 1: A method comprising: determining experimental data relating to one or more monoclonal antibodies (mAbs); determining computationally derived data relating to the one or more mAbs, where the computationally derived data comprises one or more computational parameters weighted based on an exposed area (ASA) of one or more residues of the one or more mAbs; determining a plurality of candidate predictive models based on the experimental data and the computationally derived data; determining an optimal predictive model from the plurality of candidate predictive models; and outputting the optimal predictive model.

[0181] Embodiment 2: The embodiment of embodiment 1, wherein the one or more mAbs comprise one or more of an IgG1 antibody or an IgG4 antibody. Embodiment 3: The embodiment of any one of embodiments 1-2, wherein the experimental data includes experimental viscosity data.

[0182] Embodiment 4: The embodiment of any one of embodiments 1-3, wherein the experimental viscosity data includes one or more of a dynamic viscosity value or a kinematic viscosity value. Embodiment 5: An embodiment of any of embodiments 1-4, wherein determining experimental data relating to the one or more mAbs includes measuring at least one of a dynamic viscosity value or a kinetic viscosity value based on a solution of each of the one or more mAbs and a viscometer.

[0183] Embodiment 6: An embodiment according to any of embodiments 1 to 5, wherein the computationally derived data comprises charge data associated with one or more regions associated with the sequences of the one or more mAbs, modified charge data associated with one or more regions based on the solvent accessible area of ​​residues in a homology model of the one or more mAbs, hydrophobicity index (HI), dipole moment, or isoelectric point (pI).

[0184] Embodiment 7: The embodiment of any of embodiments 1 to 6, wherein determining the computationally derived data relating to one or more mAbs comprises full antibody homology modeling of the sequence of one or more mAbs, or Fab region modeling of the antigen-binding fragment (Fab) sequence of one or more mAbs.

[0185] Embodiment 8: An embodiment of any of embodiments 1 to 7, wherein determining computationally derived data associated with the one or more mAbs comprises: determining one or more charge values ​​associated with one or more residues in one or more regions of the one or more mAbs based on a homology model of the one or more mAbs; determining a solvent accessible surface (SAS) of one or more residues in the one or more regions based on the homology model of the one or more mAbs; adjusting one or more charge values ​​associated with the one or more residues based on a weighting factor calculated using the SAS of the one or more residues relative to a total SAS associated with the one or more mAbs; and determining a charge value associated with each region of the one or more regions based on the homology model of the one or more mAbs and the adjusted one or more charge values ​​associated with the one or more residues.

[0186] Embodiment 9: An embodiment of any of embodiments 1 to 8, wherein determining a plurality of candidate predictive models based on the experimental data and the computationally derived data includes identifying one or more experimental parameters of the experimental data as dependent variables, identifying one or more computational parameters of the computationally derived data as independent variables, and determining a plurality of candidate predictive models based on the dependent variables and based on the independent variables based on a stepwise regression algorithm.

[0187] Embodiment 10: An embodiment described in any of embodiments 1 to 9, wherein determining an optimal predictive model from the plurality of candidate predictive models includes determining an Akaike Information Criterion (AIC) score for each candidate predictive model of the plurality of candidate predictive models, and determining the candidate predictive model of the plurality of candidate predictive models associated with the highest AIC score as the optimal predictive model.

[0188] Embodiment 11: The embodiment of any of embodiments 1 to 10, wherein determining an optimal predictive model from a plurality of candidate predictive models comprises determining as the optimal predictive model the candidate predictive model of the plurality of candidate predictive models associated with the smallest error in predicting the viscosity score of the mAb excluded from the experimental data and computationally derived data.

[0189] Embodiment 12: An embodiment of any of embodiments 1 to 11, further comprising receiving computationally derived data associated with the query mAb, providing the computationally derived data to a best fit predictive model, and determining a viscosity score associated with the query mAb based on the best fit predictive model.

[0190] Embodiment 13: The embodiment of embodiment 12, further comprising adjusting an appropriate formulation composition or protein engineering strategy to mitigate a particular challenge of the drug candidate under development, for example, adjusting the amount of a viscosity-lowering agent in the solution related to the query mAb, based on the viscosity score.

[0191] Embodiment 14: The embodiment of any of embodiments 1 to 13, wherein the experimental data comprises experimental aggregation data. Embodiment 15: The embodiment of embodiment 14, wherein the experimental aggregation data comprises high molecular weight (HMW) speciation data for each mAb of the one or more mAbs.

[0192] Embodiment 16: The embodiment of any of embodiments 1-15, wherein determining experimental data relating to the one or more mAbs comprises measuring the amount of HMW species formation over time based on solution and size exclusion chromatography (SEC) of each of the one or more mAbs.

[0193] Embodiment 17: The embodiment of any of embodiments 1 to 16, wherein determining experimental data relating to the one or more mAbs comprises measuring the amount of HMW species formation over time based on solution and size exclusion chromatography (SEC) of each of the one or more mAbs.

[0194] Embodiment 18: An embodiment according to any of embodiments 1 to 17, wherein the descriptor of conformational stability comprises the backbone root mean square deviation (RMSD) of the conformational structure relative to the initial structure after rigid body alignment.

[0195] Embodiment 19: The embodiment of any of embodiments 1 to 18, wherein determining computationally derived data associated with the one or more mAbs comprises one or more molecular dynamics (MD) simulations associated with the one or more mAbs.

[0196] Embodiment 20: The embodiment of any of embodiments 1 to 19, wherein determining an optimal predictive model from the plurality of candidate predictive models comprises determining as the optimal predictive model the candidate predictive model of the plurality of candidate predictive models associated with the smallest error in predicting the aggregation score of the mAb excluded from the experimental data and the computationally derived data.

[0197] Embodiment 21: An embodiment of any of embodiments 1 to 20, further comprising receiving computationally derived data related to the query mAb, providing the computationally derived data to a best fit predictive model, and determining an aggregation score based on the best fit predictive model.

[0198] Embodiment 22: The embodiment of any of embodiments 1 to 21, further comprising adjusting an appropriate formulation composition or protein engineering strategy to mitigate a particular challenge of the drug candidate under development, e.g., adjusting the amount of aggregation reducing agent in the solution associated with the query mAb, based on the aggregation score.

[0199] Embodiment 23: A method comprising receiving computationally derived data associated with a monoclonal antibody (mAb), providing the computationally derived data to a predictive model, and determining a viscosity score associated with the mAb based on the predictive model.

[0200] Embodiment 24: The embodiment of embodiment 23, further comprising adjusting an appropriate formulation composition or protein engineering strategy to alleviate a particular challenge of the drug candidate under development, for example, adjusting the amount of a viscosity-lowering agent in the solution related to the query mAb, based on the viscosity score.

[0201] Embodiment 25: The embodiment of any of embodiments 23-24, further comprising receiving sequence data associated with the mAb and determining computationally derived data based on the sequence data.

[0202] Embodiment 26: The embodiment of any one of embodiments 23 to 25, wherein the computationally derived data comprises computationally derived viscosity data. Embodiment 27: The embodiment of any one of embodiments 23 to 26, wherein the computationally derived data comprises computationally derived viscosity data.

[0203] Embodiment 28: An embodiment of any of embodiments 23 to 27, further comprising receiving computationally derived data associated with the query mAb, providing the computationally derived data to a best fit predictive model, and determining a viscosity score associated with the query mAb based on the best fit predictive model.

[0204] Embodiment 29: A method comprising receiving computationally derived data associated with a monoclonal antibody (mAb), providing the computationally derived data to a predictive model, and determining an aggregation score associated with the mAb based on the predictive model.

[0205] Embodiment 30: The embodiment of embodiment 29, further comprising adjusting an appropriate formulation composition or protein engineering strategy to alleviate a particular challenge of the drug candidate under development, e.g., adjusting the amount of an aggregation reducing agent in the solution associated with the query mAb, based on the aggregation score.

[0206] Embodiment 31: An embodiment of any of embodiments 29-30, further comprising receiving sequence data associated with the mAb and determining computationally derived data based on the sequence data.

[0207] Embodiment 32: An embodiment according to any one of embodiments 29 to 31, wherein the computationally derived data comprises computationally derived aggregation data. Embodiment 33: The embodiment of any of claims 29 to 32, wherein the computationally derived aggregation data comprises high molecular weight (HMW) speciation data for the mAb.

[0208] Embodiment 34: An embodiment of any of embodiments 29 to 33, wherein the computationally derived data comprises charge data associated with one or more regions related to the sequence of the mAb, modified charge data associated with one or more regions based on the solvent accessible areas of residues in a homology model of the mAb, hydrophobicity index (HI), dipole moment, isoelectric point (pI), aggregation propensity (AP), or a conformational stability descriptor.

[0209] Embodiment 35: An embodiment according to any of embodiments 29 to 34, further comprising determining an optimal predictive model from a plurality of candidate predictive models associated with the smallest error in predicting the aggregation score associated with the mAb.

[0210] Embodiment 36: An embodiment of any of embodiments 29 to 35, further comprising receiving computationally derived data associated with the query mAb, providing the computationally derived data associated with the query mAb to a best fit predictive model, and determining an aggregation score associated with the query mAb based on the best fit predictive model.

[0211] Although the methods and systems have been described in connection with preferred embodiments and specific examples, it is not intended that the scope be limited to the particular embodiments described, as the embodiments herein are intended in all respects to be illustrative rather than restrictive.

[0212] Unless otherwise expressly stated, it is in no way intended that any method described herein should be construed as requiring that its steps be performed in a particular order. Thus, where a claim for a method does not actually recite the order in which its steps should be followed, or where the claim or specification does not otherwise expressly state that it is limited to a particular order, no order is intended to be inferred in any respect. This holds true against all possible implicit bases for interpretation, including questions of logic regarding the placement of steps or sequence of operational flow, the apparent meaning derived from grammatical organization or punctuation, and the number or type of embodiments described herein.

[0213] Those skilled in the art will recognize, or be able to ascertain using no more than routine experimentation, many equivalents to the specific embodiments of the method and compositions described herein which equivalents are intended to be encompassed by the following claims.

Claims

1. 1. A computer-implemented method for generating a predictive model for predicting a physical property of one or more monoclonal antibodies (mAbs), comprising: determining experimental agglutination data associated with one or more monoclonal antibodies (mAbs); determining computationally derived data associated with the one or more mAbs, the computationally derived data comprising one or more computational parameters, the one or more computational parameters comprising one or more charge values ​​associated with one or more residues within one or more regions of the one or more mAbs; calculating a weighting factor based on a solvent accessible surface area (SAS) of the one or more residues in the one or more regions of the one or more mAbs; adjusting the one or more charge values ​​associated with the one or more residues based on the weighting coefficients; and determining a charge value associated with each of the one or more regions based on the adjusted one or more charge values ​​associated with the one or more residues; generating a predictive model for predicting a physical property of one or more mAbs based on the experimental agglutination data and the computationally derived data, wherein the computationally derived data includes the charge value associated with each region of the one or more regions.

2. 2. The method of claim 1, wherein the one or more mAbs comprise one or more of an IgG1 antibody or an IgG4 antibody.

3. The method of claim 1 , wherein the one or more regions are associated with a sequence of the one or more mAbs.

4. 2. The method of claim 1, wherein the experimental aggregation data comprises high molecular weight (HMW) speciation data for each mAb of the one or more mAbs.

5. 2. The method of claim 1, wherein determining the experimental aggregation data relating to the one or more mAbs comprises measuring the amount of HMW species formation over time based on solution and size exclusion chromatography (SEC) of each of the one or more mAbs.

6. 2. The method of claim 1, wherein the computationally derived data comprises charge data associated with one or more regions associated with the sequence of the one or more mAbs, modified charge data associated with the one or more regions based on the solvent accessible areas of residues in a homology model of the one or more mAbs, a hydrophobicity index (HI), a dipole moment, an isoelectric point (pI), an aggregation propensity (AP), or a conformational stability descriptor.

7. The method of claim 6, wherein the descriptor of conformational stability comprises the root mean square deviation (RMSD) of a conformational structure relative to an initial structure after rigid body alignment.

8. 2. The method of claim 1, wherein determining the computationally derived data associated with the one or more mAbs comprises one or more molecular dynamics (MD) simulations associated with the one or more mAbs.

9. 2. The method of claim 1, wherein determining the computationally derived data relating to the one or more mAbs comprises full antibody homology modeling of the sequences of the one or more mAbs, or Fab region modeling of the antigen binding fragment (Fab) sequences of the one or more mAbs.

10. calculating said weighting factor based on a solvent accessible surface area (SAS) of said one or more residues in said one or more regions of said one or more mAbs; determining the one or more charge values ​​associated with the one or more residues in the one or more regions of the one or more mAbs based on a homology model of the one or more mAbs; and determining the solvent accessible surface area (SAS) of the one or more residues in the one or more regions based on the homology model of the one or more mAbs.

11. generating the predictive model based on the experimental aggregation data and the computationally derived data; identifying one or more experimental parameters of the experimental aggregation data as dependent variables; identifying one or more calculation parameters of the computationally derived data as independent variables; and generating the predictive model based on the dependent variables and based on the independent variables based on a stepwise regression algorithm.

12. 10. The method of claim 1, further comprising determining an Akaike Information Criterion (AIC) score for the predictive model.

13. receiving computationally derived data related to a query mAb; providing the computationally derived data to the predictive model; 2. The method of claim 1, further comprising determining an aggregation score associated with the query mAb based on the predictive model.

14. 14. The method of claim 13, further comprising adjusting an amount of an aggregation reducing agent in a solution associated with the query mAb based on the aggregation score.

15. generating the predictive model based on the experimental aggregation data and the computationally derived data; generating a training data set comprising at least a portion of the experimental agglutination data and at least a portion of the computationally derived data, wherein at least the portion of the experimental agglutination data comprises direct measurements of aggregation of an antibody solution, and wherein at least the portion of the computationally derived data comprises one or more charge values ​​determined via computational modeling for residues on antibodies in the antibody solution; extracting a plurality of parameters based on the training data set; and training a machine learning based classification model configured to predict antibody aggregation based on the training dataset and the plurality of parameters.

16. An apparatus comprising: one or more processors; and a memory storing processor-executable instructions which, when executed by said one or more processors, cause said apparatus to perform a method according to any one of claims 1 to 14.

17. One or more non-transitory computer readable media storing processor executable instructions which, when executed by a processor, cause the processor to perform the method of any one of claims 1 to 14.

18. 1. A system comprising: A computing device configured to carry out the method according to any one of claims 1 to 14; a user device configured to display an output of the predictive model.

19. receiving query calculation-derived data related to a query monoclonal antibody (mAb), the calculation-derived data including one or more calculation parameters, the one or more calculation parameters including one or more charge values ​​associated with one or more residues in one or more regions of the query mAb; calculating a weighting factor based on a solvent accessible surface area (SAS) of the one or more residues in the one or more regions of the query mAb; adjusting the one or more charge values ​​associated with the one or more residues based on the weighting coefficients; and determining a charge value associated with each of the one or more regions based on the adjusted one or more charge values ​​associated with the one or more residues; providing the computationally derived data to a predictive model, the query computationally derived data including the charge value associated with each region of the one or more regions; and determining one or more of an aggregation score or a viscosity score associated with the query mAb based on the predictive model.

20. adjusting the amount of an aggregation reducing agent in the solution associated with the query mAb based on the aggregation score; or and adjusting an amount of a viscosity-lowering agent in the solution associated with the query mAb based on the viscosity score.

21. receiving sequence data related to the query mAb; The method of claim 19 , further comprising: determining the query calculation derivation data based on the sequence data.

22. The method of claim 19 , wherein the query calculation-derived data comprises calculation-derived aggregated data.

23. 23. The method of claim 22, wherein the computationally derived aggregation data comprises high molecular weight (HMW) speciation data for the mAb.

24. The method of claim 19 , wherein the query calculation-derived data comprises calculation-derived viscosity data.

25. 25. The method of claim 24, wherein the query calculation derived viscosity data includes one or more of a dynamic viscosity value or a kinematic viscosity value.

26. 20. The method of claim 19, wherein the computationally derived data comprises charge data associated with one or more regions associated with the sequence of the mAb, modified charge data associated with the one or more regions based on the solvent accessible areas of residues in a homology model of the mAb, a hydrophobicity index (HI), a dipole moment, an isoelectric point (pI), an aggregation propensity (AP), or a conformational stability descriptor.

27. calculating said weighting factor based on a solvent accessible surface area (SAS) of said one or more residues in said one or more regions of said one or more mAbs; determining the one or more charge values ​​associated with the one or more residues in the one or more regions of the one or more mAbs based on a homology model of the one or more mAbs; and determining the solvent accessible surface area (SAS) of the one or more residues in the one or more regions based on the homology model of the one or more mAbs.

28. The predictive model is based on experimental aggregation data and computationally derived data, identifying one or more experimental parameters of the experimental aggregation data as dependent variables; identifying one or more calculation parameters of the computationally derived data as independent variables; generating the predictive model based on the dependent variables and based on the independent variables based on a stepwise regression algorithm; 20. The method of claim 19, produced according to a method comprising:

29. The predictive model is based on experimental aggregation data and computationally derived data, generating a training data set comprising at least a portion of the experimental agglutination data and at least a portion of the computationally derived data, wherein at least the portion of the experimental agglutination data comprises direct measurements of aggregation of an antibody solution, and wherein at least the portion of the computationally derived data comprises one or more charge values ​​determined via computational modeling for residues on antibodies in the antibody solution; extracting a plurality of parameters based on the training data set; training a machine learning based classification model configured to predict antibody aggregation based on the training dataset and the plurality of parameters; 20. The method of claim 19, produced according to a method comprising:

30. The predictive model is based on experimental viscosity data and computationally derived data, identifying one or more experimental parameters of the experimental viscosity data as dependent variables; identifying one or more calculation parameters of the computationally derived data as independent variables; generating the predictive model based on the dependent variables and based on the independent variables based on a stepwise regression algorithm; 20. The method of claim 19, produced according to a method comprising:

31. The predictive model is based on experimental viscosity data and computationally derived data, generating a training data set comprising at least a portion of the experimental viscosity data and at least a portion of the computationally derived data, wherein at least the portion of the experimental viscosity data comprises direct measurements of viscosity of an antibody solution and wherein at least the portion of the computationally derived data comprises one or more charge values ​​determined via computational modeling for residues on antibodies of the antibody solution; extracting a plurality of parameters based on the training data set; training a machine learning based classification model configured to predict the viscosity of an antibody based on the training dataset and the plurality of parameters; 20. The method of claim 19, produced according to a method comprising:

32. An apparatus comprising: one or more processors; and a memory storing processor-executable instructions which, when executed by said one or more processors, cause said apparatus to perform a method according to any one of claims 19 to 31.

33. One or more non-transitory computer readable media storing processor executable instructions which, when executed by a processor, cause the processor to perform the method of any one of claims 19 to 31.

34. 1. A system comprising: A computing device configured to carry out the method according to any one of claims 19 to 31; a user device configured to display an output of the predictive model.

Images