Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Optimization of therapeutic antibodies for reduced self-association and non-specific binding via interpretable machine learning



Antibody development, delivery, and efficacy are influenced by antibody-antigen affinity interactions, off-target interactions that reduce antibody bioavailability and pharmacokinetics, and repulsive self-interactions that increase the stability of concentrated antibody formulations and reduce their corresponding viscosity. Yet identifying antibody variants with optimal combinations of these three types of interactions is challenging. Here we show that interpretable machine-learning classifiers, leveraging antibody structural features descriptive of their variable regions and trained on experimental data for a panel of 80 clinical-stage monoclonal antibodies, can identify antibodies with optimal combinations of low off-target binding in a common physiological-solution condition and low self-association in a common antibody-formulation condition. For three clinical-stage antibodies with suboptimal combinations of off-target binding and self-association, the classifiers predicted variable-region mutations that optimized non-affinity interactions while maintaining high-affinity antibody-antigen interactions. Interpretable machine-learning models may facilitate the optimization of antibody candidates for therapeutic applications.

This is a preview of subscription content, access via your institution

Access options

Rent or buy this article

Prices vary by article type



Prices may be subject to local taxes which are calculated during checkout

Fig. 1: Approach for predicting non-affinity interactions for clinical-stage antibodies.
Fig. 2: Experimental classification of clinical-stage antibodies with different levels of self-association and non-specific binding.
Fig. 3: Interpretable models for predicting the levels of self-association and non-specific binding for clinical-stage antibodies.
Fig. 4: Prediction of mutations that co-optimize affinity and non-affinity interactions for three clinical-stage antibodies.
Fig. 5: Predicted mutations in clinical-stage antibodies display co-optimized levels of affinity and non-affinity interactions.

Data availability

The main data supporting the results of this study are available within the paper and its Supplementary Information. Source data for Fig. 5 are provided with this paper.

Code availability

The scripts to generate the main figures and to perform the analysis described in this paper are available at the Tessier lab GitHub repository at


  1. Jain, T. et al. Biophysical properties of the clinical-stage antibody landscape. Proc. Natl Acad. Sci. USA 114, 944–949 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  2. Makowski, E. K., Wu, L., Gupta, P. & Tessier, P. M. Discovery-stage identification of drug-like antibodies using emerging experimental and computational methods. mAbs 13, 1895540 (2021).

  3. Labrijn, A. F., Janmaat, M. L., Reichert, J. M. & Parren, P. Bispecific antibodies: a mechanistic review of the pipeline. Nat. Rev. Drug Discov. 18, 585–608 (2019).

    Article  CAS  PubMed  Google Scholar 

  4. Shim, H. Bispecific antibodies and antibody-drug conjugates for cancer therapy: technological considerations. Biomolecules 10, 360 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  5. Drago, J. Z., Modi, S. & Chandarlapaty, S. Unlocking the potential of antibody–drug conjugates for cancer therapy. Nat. Rev. Clin. Oncol. 18, 327–344 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

  6. Dean, A. Q., Luo, S., Twomey, J. D. & Zhang, B. Targeting cancer with antibody-drug conjugates: promises and challenges. mAbs 13, 1951427 (2021).

  7. Carter, P. J. Potent antibody therapeutics by design. Nat. Rev. Immunol. 6, 343–357 (2006).

    Article  CAS  PubMed  Google Scholar 

  8. Carter, P. J. & Rajpal, A. Designing antibodies as therapeutics. Cell 185, 2789–2805 (2022).

    Article  CAS  PubMed  Google Scholar 

  9. Leavy, O. Therapeutic antibodies: past, present and future. Nat. Rev. Immunol. 10, 297 (2010).

    Article  CAS  PubMed  Google Scholar 

  10. Chames, P., Van Regenmortel, M., Weiss, E. & Baty, D. Therapeutic antibodies: successes, limitations and hopes for the future. Br. J. Pharmacol. 157, 220–233 (2009).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  11. Makowski, E. K. et al. Co-optimization of therapeutic antibody affinity and specificity using machine learning models that generalize to novel mutational space. Nat. Commun. 13, 3788 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  12. Gupta, P. et al. Antibodies with weakly basic isoelectric points minimize trade-offs between formulation and physiological colloidal properties. Mol. Pharm. 19, 775–787 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  13. Rabia, L. A., Desai, A. A., Jhajj, H. S. & Tessier, P. M. Understanding and overcoming trade-offs between antibody affinity, specificity, stability and solubility. Biochem. Eng. J. 137, 365–374 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  14. Kingsbury, J. S. et al. A single molecular descriptor to predict solution behavior of therapeutic antibodies. Sci. Adv. 6, eabb0372 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  15. Starr, C. G. et al. Ultradilute measurements of self-association for the identification of antibodies with favorable high-concentration solution properties. Mol. Pharm. 18, 2744–2753 (2021).

    Article  CAS  PubMed  Google Scholar 

  16. Makowski, E. K., Wu, L., Desai, A. A. & Tessier, P. M. Highly sensitive detection of antibody nonspecific interactions using flow cytometry. mAbs 13, 1951426 (2021).

  17. Xu, Y. et al. Addressing polyspecificity of antibodies selected from an in vitro yeast presentation system: a FACS-based, high-throughput selection and analytical tool. Protein Eng. Des. Sel. 26, 663–670 (2013).

    Article  CAS  PubMed  Google Scholar 

  18. Ahmed, L. et al. Intrinsic physicochemical profile of marketed antibody-based biotherapeutics. Proc. Natl Acad. Sci. USA 118, e2020577118 (2021).

  19. Yadav, S., Laue, T. M., Kalonia, D. S., Singh, S. N. & Shire, S. J. The influence of charge distribution on self-association and viscosity behavior of monoclonal antibody solutions. Mol. Pharm. 9, 791–802 (2012).

    Article  CAS  PubMed  Google Scholar 

  20. Yadav, S. et al. Establishing a link between amino acid sequences and self-associating and viscoelastic behavior of two closely related monoclonal antibodies. Pharm. Res. 28, 1750–1764 (2011).

    Article  CAS  PubMed  Google Scholar 

  21. Xolair. Prescribing Information (Genentech Inc., 2021).

  22. Dyson, M. R. et al. Beyond affinity: selection of antibody variants with optimal biophysical properties and reduced immunogenicity from mammalian display libraries. mAbs 12, 1829335 (2020).

  23. Wang, N. et al. Opalescence of an IgG1 monoclonal antibody formulation is mediated by ionic strength and excipients. Biopharm Int. 22, 36–47 (2009).

  24. Salinas, B. A. et al. Understanding and modulating opalescence and viscosity in a monoclonal antibody formulation. J. Pharm. Sci. 99, 82–93 (2010).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  25. Goldberg, D. S., Bishop, S. M., Shah, A. U. & Sathish, H. A. Formulation development of therapeutic monoclonal antibodies using high-throughput fluorescence and static light scattering techniques: role of conformational and colloidal stability. J. Pharm. Sci. 100, 1306–1315 (2011).

    Article  CAS  PubMed  Google Scholar 

  26. Shi, G. H. et al. Subcutaneous injection site pain of formulation matrices. Pharm. Res. 38, 779–793 (2021).

    Article  CAS  PubMed  Google Scholar 

  27. Chabra, S. et al. Ixekizumab citrate-free formulation: results from two clinical trials. Adv. Ther. 39, 2862–2872 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  28. Grinshpun, B. et al. Identifying biophysical assays and in silico properties that enrich for slow clearance in clinical-stage therapeutic antibodies. mAbs 13, 1932230 (2021).

  29. Hötzel, I. et al. A strategy for risk mitigation of antibodies with fast clearance. mAbs 4, 753–760 (2012).

  30. Neergaard, M. S., Nielsen, A. D., Parshad, H. & Van De Weert, M. Stability of monoclonal antibodies at high-concentration: head-to-head comparison of the IgG1 and IgG4 subclass. J. Pharm. Sci. 103, 115–127 (2014).

    Article  CAS  PubMed  Google Scholar 

  31. Lai, P. K. et al. Differences in human IgG1 and IgG4 S228P monoclonal antibodies viscosity and self-interactions: experimental assessment and computational predictions of domain interactions. mAbs 13, 1991256 (2021).

  32. Sickmier, E. A. et al. The panitumumab EGFR complex reveals a binding mechanism that overcomes cetuximab induced resistance. PLoS ONE 11, e0163366 (2016).

    Article  PubMed  PubMed Central  Google Scholar 

  33. Bohrmann, B. et al. Gantenerumab: a novel human anti-Aβ antibody demonstrates sustained cerebral amyloid-β binding and elicits cell-mediated removal of human amyloid-β. J. Alzheimers Dis. 28, 49–69 (2012).

    Article  CAS  PubMed  Google Scholar 

  34. Weihofen, A. et al. Development of an aggregate-selective, human-derived α-synuclein antibody BIIB054 that ameliorates disease phenotypes in Parkinson’s disease models. Neurobiol. Dis. 124, 276–288 (2019).

    Article  CAS  PubMed  Google Scholar 

  35. De Groot, A. S. & Martin, W. Reducing risk, improving outcomes: bioengineering less immunogenic protein therapeutics. Clin. Immunol. 131, 189–201 (2009).

    Article  PubMed  Google Scholar 

  36. De Groot, A. S. & Scott, D. W. Immunogenicity of protein therapeutics. Trends Immunol. 28, 482–490 (2007).

    Article  PubMed  Google Scholar 

  37. Apgar, J. R. et al. Modeling and mitigation of high-concentration antibody viscosity through structure-based computer-aided protein design. PLoS ONE 15, e0232713 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  38. Tomar, D. S. et al. In-silico prediction of concentration-dependent viscosity curves for monoclonal antibody solutions. mAbs 9, 476–489 (2017).

  39. Boughter, C. T. et al. Biochemical patterns of antibody polyreactivity revealed through a bioinformatics-based analysis of CDR loops. eLife 9, e61393 (2020).

  40. Lai, P. K., Gallegos, A., Mody, N., Sathish, H. A. & Trout, B. L. Machine learning prediction of antibody aggregation and viscosity for high concentration formulation development of protein therapeutics. mAbs 14, 2026208 (2022).

  41. Han, X., Shih, J., Lin, Y., Chai, Q. & Cramer, S. M. Development of QSAR models for in silico screening of antibody solubility. mAbs 14, 2062807 (2022).

  42. Mason, D. M. et al. Optimization of therapeutic antibodies by predicting antigen specificity from antibody sequence via deep learning. Nat. Biomed. Eng. 5, 600–612 (2021).

    Article  CAS  PubMed  Google Scholar 

  43. Saka, K. et al. Antibody design using LSTM based deep generative model from phage display library for affinity maturation. Sci. Rep. 11, 5852 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  44. Hie, B. L. et al. Efficient evolution of human antibodies from general protein language models and sequence information alone. Nat. Biotechnol. (2023).

  45. Shin, J. E. et al. Protein design and variant prediction using autoregressive generative models. Nat. Commun. 12, 2403 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  46. Neal, B. L., Asthagiri, D. & Lenhoff, A. M. Molecular origins of osmotic second virial coefficients of proteins. Biophys. J. 75, 2469–2477 (1998).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  47. Lomakin, A., Asherie, N. & Benedek, G. B. Aeolotopic interactions of globular proteins. Proc. Natl Acad. Sci. USA 96, 9465–9468 (1999).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  48. Elcock, A. H., Sept, D. & McCammon, J. A. Computer simulation of protein−protein interactions. J. Phys. Chem. B 105, 1504–1518 (2001).

    Article  CAS  Google Scholar 

  49. Sharma, V. K. et al. In silico selection of therapeutic antibodies for development: viscosity, clearance, and chemical stability. Proc. Natl Acad. Sci. USA 111, 18601–18606 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  50. Raybould, M. I. J. et al. Five computational developability guidelines for therapeutic antibody profiling. Proc. Natl Acad. Sci. USA 116, 4025–4030 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  51. Negron, C., Fang, J., McPherson, M. J., Stine, W. B. Jr. & McCluskey, A. J. Separating clinical antibodies from repertoire antibodies, a path to in silico developability assessment. mAbs 14, 2080628 (2022).

  52. Connolly, B. D. et al. Weak interactions govern the viscosity of concentrated antibody solutions: high-throughput analysis using the diffusion interaction parameter. Biophys. J. 103, 69–78 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  53. Kelly, R. L. et al. Chaperone proteins as single component reagents to assess antibody nonspecificity. mAbs 9, 1036–1040 (2017).

  54. Datta-Mannan, A. et al. The interplay of non-specific binding, target-mediated clearance and FcRn interactions on the pharmacokinetics of humanized antibodies. mAbs 7, 1084–1093 (2015).

  55. Chung, S. et al. An in vitro FcRn- dependent transcytosis assay as a screening tool for predictive assessment of nonspecific clearance of antibody therapeutics in humans. mAbs 11, 942–955 (2019).

  56. Liu, Y. et al. High-throughput screening for developability during early-stage antibody discovery using self-interaction nanoparticle spectroscopy. mAbs 6, 483–492 (2014).

  57. Sule, S. V., Dickinson, C. D., Lu, J., Chow, C. K. & Tessier, P. M. Rapid analysis of antibody self-association in complex mixtures using immunogold conjugates. Mol. Pharm. 10, 1322–1331 (2013).

    Article  CAS  PubMed  Google Scholar 

  58. Sun, T. et al. High throughput detection of antibody self-interaction by bio-layer interferometry. mAbs 5, 838–841 (2013).

  59. Mouquet, H. et al. Polyreactivity increases the apparent affinity of anti-HIV antibodies by heteroligation. Nature 467, 591–595 (2010).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  60. Wardemann, H. et al. Predominant autoantibody production by early human B cell precursors. Science 301, 1374–1377 (2003).

    Article  CAS  PubMed  Google Scholar 

  61. Jacobs, S. A., Wu, S. J., Feng, Y., Bethea, D. & O’Neil, K. T. Cross-interaction chromatography: a rapid method to identify highly soluble monoclonal antibody candidates. Pharm. Res. 27, 65–71 (2010).

    Article  CAS  PubMed  Google Scholar 

  62. Estep, P. et al. An alternative assay to hydrophobic interaction chromatography for high-throughput characterization of monoclonal antibodies. mAbs 7, 553–561 (2015).

  63. Haverick, M., Mengisen, S., Shameem, M. & Ambrogelly, A. Separation of mAbs molecular variants by analytical hydrophobic interaction chromatography HPLC: overview and applications. mAbs 6, 852–858 (2014).

  64. Phan, S., Walmer, A., Shaw, E. W. & Chai, Q. High-throughput profiling of antibody self-association in multiple formulation conditions by PEG stabilized self-interaction nanoparticle spectroscopy. mAbs 14, 2094750 (2022).

  65. Liberis, E., Velickovic, P., Sormanni, P., Vendruscolo, M. & Liò, P. Parapred: antibody paratope prediction using convolutional and recurrent neural networks. Bioinformatics 34, 2944–2950 (2018).

    Article  CAS  PubMed  Google Scholar 

  66. Bailly, M. et al. Predicting antibody developability profiles through early stage discovery screening. mAbs 12, 1743053 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  67. Pepinsky, R. B. et al. Improving the solubility of anti-LINGO-1 monoclonal antibody Li33 by isotype switching and targeted mutagenesis. Protein Sci. 19, 954–966 (2010).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  68. Heads, J. T. et al. Electrostatic interactions modulate the differential aggregation propensities of IgG1 and IgG4P antibodies and inform charged residue substitutions for improved developability. Protein Eng. Des. Sel. 32, 277–288 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  69. Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  70. Molecular Operating Environment (Chemical Computing Group Inc., 2020).

  71. Desai, A. A., Zupancic, J. M., Smith, M. D. & Tessier, P. M. Isolating anti-amyloid antibodies from yeast-displayed libraries. Methods Mol. Biol. 2491, 471–490 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

  72. De Groot, A. S. et al. Better epitope discovery, precision immune engineering, and accelerated vaccine design using immunoinformatics tools. Front. Immunol. 11, 442 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  73. Jawa, V. et al. T-cell dependent immunogenicity of protein therapeutics pre-clinical assessment and mitigation—updated consensus and review 2020. Front. Immunol. 11, 1301 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  74. Moise, L. et al. Universal H1N1 influenza vaccine development: identification of consensus class II hemagglutinin and neuraminidase epitopes derived from strains circulating between 1980 and 2011. Hum. Vaccin. Immunother. 9, 1598–1607 (2013).

    Article  CAS  PubMed  Google Scholar 

Download references


We thank members of the Tessier lab for helpful suggestions. This work was supported by the Biomolecular Interaction Technology Center, the National Institutes of Health (R35GM136300 to P.M.T., 1T32GM140223-01 to E.K.M., and F32GM137513 to J.S.S.), the National Science Foundation (CBET 1813963, 1605266 and 1804313 to P.M.T.) and the Albert M. Mattocks Chair (to P.M.T.).

Author information

Authors and Affiliations



E.K.M. developed the methodology, with conceptual input from P.M.T. E.K.M., T.W., J.M.Z., L.W., J.H. and J.S.S. performed antibody synthesis and preparation. E.K.M. and J.M.Z. performed antibody characterization. A.D.G., W.D.M. and S.L.E. designed the immunogenicity analysis. E.K.M. performed structural modelling and the computational analyses of antibody sequences. E.K.M. and P.M.T. wrote and revised the manuscript.

Corresponding author

Correspondence to Peter M. Tessier.

Ethics declarations

Competing interests

The University of Michigan and Sanofi have filed a patent application for the CS-SINS method (WO 2022/148777 A1 (2022)). P.M.T. is a member of the scientific advisory board for Nabla Bio, and has received honoraria for invited presentations including on this research from GlaxoSmithKline, Bristol Myers Squibb, Janssen and Genentech. The other authors declare no competing interests.

Peer review

Peer review information

Nature Biomedical Engineering thanks Charlotte Deane, Sai Reddy and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 Correlation analysis between structure-based antibody Fv molecular features and self-association and non-specific binding measurements.

(A) Molecular features from Fv homology models, including hydrophobic (hyd.), positively charged (pos.), negatively charged (neg.), and ionic (positively and negatively charged) features, were found to be correlated to different extents with the self-association (CS-SINS) and non-specific binding (PSP) measurements for the 80 clinical-stage antibodies in this study. (B) Relationship between the CS-SINS and PSP measurements and a subset of key molecular features. In (A), total area features are those that describe the total additive patch area in Å, the largest patch area features are those that describe the area of the single largest patch in Å, the # of patches features are those that describe the additive number of patches larger than 50 Å2, and the % surface area features are those that describe the total patch area relative to the total surface area of the Fv. Moreover, hyd. stands for hydrophobic, hph. stands for hydrophilic, vdw stands for van der Waals, and ASA stands for accessible surface area. In (A-B), independent two-sided t-tests were performed to determine significance and the p-values are < 0.05 (*), < 0.01 (**) and < 0.001 (***).

Extended Data Fig. 2 Application of models to previously reported antibodies with diverse physicochemical properties.

(A) Variants of an antibody, AB-001, with low viscosity (150 mg/mL, 10 mM histidine 5% sucrose buffer, pH 5.8) are well discriminated from variants that display high viscosity using our self-association model (Apgar et al., PLoS One, 2020). (B) Variants of bococizumab with low non-specific binding (as judged by binding to bovine serum albumin) are identified using our non-specific binding model (Dyson et al., mAbs, 2020). (C-D) Variants of emibetuzumab with reduced non-specific binding and modestly increased self-association are correctly described in terms of the direction of the predictions, as the models predict (D) reduced non-specific binding and (C) modest increased in self-association without promoting high self-association (Makowski et al., Nat Commun, 2022). (E) Variants of omalizumab with reduced viscosity (125 mg/mL; 15 mM histidine, pH 6.0) include those that are predicted near the decision boundary for low self-association. In (A), the viscosity measurements are indicated by marker size (larger size indicates higher viscosity). In (C-E), the measurements for the corresponding properties are reported next to the data points, including the (C) PSP scores, (D) CS-SINS scores, and (E) viscosities. Model features are normalized between 0 and 1 for optimized model fitting.

Supplementary information

Main Supplementary Information

Supplementary figures.

Reporting Summary

Supplementary Dataset 1

Summary of the clinical-stage antibodies in this study.

Supplementary Dataset 2

Summary of CS-SINS and PSP data for 80 clinical-stage antibodies.

Supplementary Dataset 3

Description of the 33 features extracted from MOE, with descriptions.

Supplementary Dataset 4

Summary of the calculated features values used in the self-association and non-specific binding models for each antibody.

Supplementary Dataset 5

Summary of model performance for the self-association and non-specific binding models trained using “3*” or “3” features, as explained in Supplementary Fig. 3.

Supplementary Dataset 6

Summary of the impact of single mutations for cinpanemab, gantenerumab and panitumumab on the five antibody features used in the self-association and non-specific binding models.

Source data

Source Data for Fig. 5

CS-SINS, PSP and EC50 measurements for the parental antibodies and variants.

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and Permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Makowski, E.K., Wang, T., Zupancic, J.M. et al. Optimization of therapeutic antibodies for reduced self-association and non-specific binding via interpretable machine learning. Nat. Biomed. Eng (2023).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI:


Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing