HLAIIPred

1 Introduction

ImmunoSeq¹ is a deep learning model based on the Transformer architecture, trained on mass spectrometry–identified HLA-II eluted peptide datasets, and designed to predict HLA-II binding peptides presented on the surface of antigen-presenting cells. By learning pMHC interaction patterns, the model can identify the core residues that determine binding and count the number of effective core peptides, yielding a quantitative metric, HLAIIPred_count, for assessing the potential immunogenicity of antibodies. A lower HLAIIPred_count indicates that fewer core peptides in the antibody sequence are effectively presented by HLA-II, corresponding to a lower risk of inducing anti-drug antibody (ADA) generation; a higher value suggests a greater immunogenicity risk. The tool also supports residue-level localization of immunogenicity hotspots within the sequence, facilitating the identification of key sites that may trigger immune responses and providing a basis for immunogenicity assessment and sequence optimization.

Its core function is to count immunogenicity core peptides for an input antibody protein sequence and thereby assess the antibody's immunogenicity risk. On this basis, users can further combine residue-level hotspot distribution to analyze the contributions of the heavy chain, light chain, and individual domains to immunogenicity, assisting in antibody humanization design, candidate molecule screening, and developability assessment.

Figure 1. Model framework of HLAIIPred.

Figure 2. Binary classification metrics of HLAIIPred_count and therapeutic antibody ADA%.

Figure 3. Correlation plot between HLAIIPred_count and ADA% of therapeutic antibodies.

2 Parameters

• Heavy Chain Sequence：Fv protein sequence of antibody heavy chain.
• Light Chain Sequence：Fv protein sequence of antibody light chain.

3 Results Explanation

• HLAIIPred_count: Number of immunogenicity core peptides. A lower value indicates a lower risk of anti-drug antibody (ADA) generation.
• HLAIIPred_count distribution of therapeutic antibodies from different origins
  The figure presents the HLAIIPred_count prediction results for the input antibody and allows intuitive assessment of which source of therapeutic antibody it is most similar to, including Human, Humanized/Chimeric, Humanized, Chimeric, and Mouse.

Figure 4. HLAIIPred_count distribution of therapeutic antibodies from different origins.

• Immunogenicity hotspot region plot
  The figure shows the immunogenicity hotspot regions of the input antibody heavy chain (top) and light chain (bottom); the darker the color, the higher the probability that the region is predicted to be an immunogenicity hotspot.

Figure 4. Example plot of immunogenicity hotspot regions.

• Residue-level Immunogenicity Scores and Characteristic Motifs Summary：
  The table lists the immunogenicity score of each amino acid position and its corresponding motif in the input antibody heavy chain (top) and light chain (bottom).

Tabel 1. 输入抗体的免疫原性热点区域图

4 Reference

[1] Kovaltsuk, A. et al. Observed Antibody Space: A Resource for Data Mining Next-Generation Sequencing of Antibody Repertoires. The Journal of Immunology 201, 2502–2509. https://doi.org/10.4049/jimmunol.1800708