Immunopeptidomic Data Integration to Artificial Neural Networks Enhances Protein-Drug Immunogenicity Prediction

Abstract

Recombinant DNA technology has, in the last decades, contributed to a vast expansion of the use of protein drugs as pharmaceutical agents. However, such biological drugs can lead to the formation of anti-drug antibodies (ADAs) that may result in adverse effects, including allergic reactions and compromised therapeutic efficacy. Production of ADAs is most often associated with activation of CD4 T cell responses resulting from proteolysis of the biotherapeutic and loading of drug-specific peptides into major histocompatibility complex (MHC) class II on professional antigen-presenting cells. Recently, readouts from MHC-associated peptide proteomics (MAPPs) assays have been shown to correlate with the presence of CD4 T cell epitopes. However, the limited sensitivity of MAPPs challenges its use as an immunogenicity biomarker. In this work, MAPPs data was used to construct an artificial neural network (ANN) model for MHC class II antigen presentation. Using Infliximab and Rituximab as showcase stories, the model demonstrated an unprecedented performance for predicting MAPPs and CD4 T cell epitopes in the context of protein-drug immunogenicity, complementing results from MAPPs assays and outperforming conventional prediction models trained on binding affinity data.

Improved prediction of HLA antigen presentation hotspots: Applications for immunogenicity risk assessment of therapeutic proteins

“…To further test the developed tool, we evaluated its power to predict the outcome of a series of MAPPs experiments of therapeutic proteins and compared the performance with that of NetMHCIIpan‐3.2 and MixMHC2pred. Therapeutic proteins included were vatreptacog alfa (coagulation factor VII analogue), coagulation factor X analogue (referred to as factor X), liraglutide peptide backbone (GLP‐1 agonist) and infliximab (data obtained from Barra et al 32 and Karle et al 35 ).…”

“…As a further benchmark evaluation, we next compared NNAlign_MA with two earlier methods for prediction of MHC class II antigen presentation developed by us. The first method (termed Barra) was described in Barra et al 32 and consists of a model trained on a limited set of MS EL data and benchmarked on MAPPs ligands and T‐cell epitope responses in infliximab. The second model is the most recent version of NetMHCIIpan (version 4.0) 29,36 .…”

“…As it is a labour‐intense, low‐throughput and costly assay, the industry would benefit greatly from an in silico tool that could predict presented MHC‐II peptides. Barra et al 32 recently demonstrated that in silico models trained on a small MS MHC‐II elution ligand data set could predict the outcome of MAPPs assays on therapeutic proteins. However, this study was limited by a small donor cohort and a very limited set of investigated proteins, which complicated the performance evaluation.…”

“…Recently, Barra et al 32 illustrated how neural networks could be trained on MS data, which could be effectively applied to predict HLA‐DR antigen presentation hotspots and T‐cell immunogenicity for infliximab and rituximab. This method showed superior performance to conventional HLA binding prediction methods trained on in vitro BA data.…”

Toward Drug-Like Multispecific Antibodies by Design

“…Without that critical piece of information, it is difficult to develop a model for MHC peptide binding. To solve this problem, a variety of innovative approaches have been explored, the most recent of which is the development of artificial neural networks that can simultaneously analyze the mass spectrometry data while developing individual MHC binding models [ 257 , 260 ]. The authors report that this strategy results in an immunogenicity predictor that met or exceeded the predictive capacity of previous methods in their test system.…”