Humanization

1 Introduction

Monoclonal antibody therapeutics typically originate from non-human sources (usually mice), which may trigger immune responses in humans. Antibody humanization aims to modify the variable region sequences of antibodies to obtain antibodies that do not elicit immune responses. We utilized nearly one billion antibody sequences from the OAS^1,2 database to establish an antibody humanness evaluation AI model capable of distinguishing between human and non-human antibody variable region sequences. The scores output by the model are negatively correlated with the experimental immunogenicity (ADA) of existing FDA-approved antibody therapies³. Following the approach of Marks and Hummer³, we combined this model with a Beam Search algorithm to develop an antibody sequence humanization tool. This tool aims to maximize the level of humaness of antibodies while minimizing number of mutations and maintaining key characteristics such as affinity, thereby reducing their immunogenicity.

Table 1. Metrics of the Humanness Evaluation Model on the OAS Dataset.

Figure 1. Confusion Matrix of the Humanness Evaluation Model on the OAS Test Set.

Figure 2. The humanness evaluation model scores are found to be negatively correlated with the fraction of patients with observed immunogenic responses, which was obtained from FDA labels of approved antibody therapeutics and clinical studies of therapeutics still in clinical trials.

2 Parameters

• Fv Sequence: amino acid sequence of an antibody Fv region in FASTA format. If the input sequence contains parts outside the V region, the tool will automatically identify and extract the V region.
• K：int, k represents the number of candidate antibody sequences with a high level of humanization that the user expects to be designed from the parent antibody.
• Score Thres：float, the score of the humanized antibody should be higher than score_thres.
• Dont Mutate Cdr: bool, default value is 'True'，when the value is 'True', it indicates that mutations in the CDR region are not allowed.

3 Results Explanation

The returned result is a table formatted as follows.

Table 2. Format of the returned table.

The first row represents the user-input parent antibody, and the remaining k rows are the k humanized antibodies designed by the Humanization tool. The meanings of each field are as follows:

• Fv Sequence：The Fv sequence of the antibody;
• Humanness Score：The score from the antibody humanness evaluation tool, ranging from 0 to 1, where a higher score indicates a greater degree of humanization;
• Mutations：The mutations that occurred in the humanized antibodies compared to the parent antibody；

4 Reference

[1] Olsen TH, Boyles F, Deane CM. Observed Antibody Space: A diverse database of cleaned, annotated, and translated unpaired and paired antibody sequences. Protein Sci. 2022 Jan;31(1):141-146. doi: 10.1002/pro.4205. Epub 2021 Oct 29. PMID: 34655133; PMCID: PMC8740823. https://doi.org/10.1002/pro.4205
[2] Aleksandr Kovaltsuk, Jinwoo Leem, Sebastian Kelm, James Snowden, Charlotte M. Deane, Konrad Krawczyk; Observed Antibody Space: A Resource for Data Mining Next-Generation Sequencing of Antibody Repertoires. J Immunol 15 October 2018; 201 (8): 2502–2509. https://doi.org/10.4049/jimmunol.1800708
[3] Claire Marks, Alissa M Hummer, Mark Chin, Charlotte M Deane, Humanization of antibodies using a machine learning approach on large-scale repertoire data, Bioinformatics, Volume 37, Issue 22, November 2021, Pages 4041–4047, https://doi.org/10.1093/bioinformatics/btab434