PANDORA: A Fast, Anchor-Restrained Modelling Protocol for Peptide: MHC Complexes

Abstract

Deeper understanding of T-cell-mediated adaptive immune responses is important for the design of cancer immunotherapies and antiviral vaccines against pandemic outbreaks. T-cells are activated when they recognize foreign peptides that are presented on the cell surface by Major Histocompatibility Complexes (MHC), forming peptide:MHC (pMHC) complexes. 3D structures of pMHC complexes provide fundamental insight into T-cell recognition mechanism and aids immunotherapy design. High MHC and peptide diversities necessitate efficient computational modelling to enable whole proteome structural analysis. We developed PANDORA, a generic modelling pipeline for pMHC class I and II (pMHC-I and pMHC-II), and present its performance on pMHC-I here. Given a query, PANDORA searches for structural templates in its extensive database and then applies anchor restraints to the modelling process. This restrained energy minimization ensures one of the fastest pMHC modelling pipelines so far. On a set of 835 pMHC-I complexes over 78 MHC types, PANDORA generated models with a median RMSD of 0.70 Å and achieved a 93% success rate in top 10 models. PANDORA performs competitively with three pMHC-I modelling state-of-the-art approaches and outperforms AlphaFold2 in terms of accuracy while being superior to it in speed. PANDORA is a modularized and user-configurable python package with easy installation. We envision PANDORA to fuel deep learning algorithms with large-scale high-quality 3D models to tackle long-standing immunology challenges.

HLA3DB: comprehensive annotation of peptide/HLA complexes enables blind structure prediction of T cell epitopes

“…Next, we compare RepPred with six existing approaches for pHLA-I structural modeling: GradDock 24 , APE-Gen 33 , a method by Keller, et al 29 , PANDORA 25 , AlphaFold2 34 , and a peptide/MHC fine-tuned version of AlphaFold (AF-FT) 35 . We focus our comparison on HLA-A02 targets and find that RepPred outperforms five of the six methods by at least 32% with respect to D-score (Fig.…”

“…Finally, we combine our basis set of distinct structural templates with a regression model trained on Rosetta energy terms to develop a structural modeling approach for nonamer/HLA complexes (RepPred). Using a cross-validation technique, we find that our method outperforms six state-of-the-art methods 24,25,29,[33][34][35] showing a 19% accuracy improvement relative to the top method 35 , while consistently identifying the correct templates for target backbones that sparsely populated in the PDB.…”

“…Constrained backbone based predictions assume that all backbones of the same length are similar in their conformation 12,[16][17][18][19][20][21] . On the other hand, constrained termini methods offer a less restrictive view as they assume that termini residues are confined to defined "pockets" in the peptide-binding groove while the locations of the other peptide residues are said to be variable [22][23][24][25][26][27] . Lastly, fragment-based docking strategies utilize minimization protocols to sample the peptide backbone ab initio, allowing for a greater degree of flexibility 28,29 .…”

Accurate modeling of peptide-MHC structures with AlphaFold

“…Protein modeling tools such as Rosetta [2] and MODELLER [3] were adapted to pMHC structure prediction in the context of immunogenicity prediction [4] and the modeling of pMHC-TCR complexes [5]. An example of an automated user-friendly tool is PANDORA [6], a MODELLER-based pipeline. Since the deep learning revolution in protein structure prediction [7,8], AlphaFold [8,9] (AF) has been applied to pMHC structure and binding prediction [10], and custom neural nets were built specifically for this task [11,12].…”

“…TFold works for peptides of different lengths in class I and class II structures with MHC alleles from human, mouse, and a few other species. It demonstrates high accuracy, for class I pMHCs significantly outperforming PANDORA [6]. For class II pMHCs, it outperforms state-of-the-art methods netMHCIIpan 3.2 and 4.0 [13,14] in peptide register prediction.…”