Artificial intelligence applied in neoantigen identification facilitates personalized cancer immunotherapy

Abstract

The field of cancer neoantigen investigation has developed swiftly in the past decade. Predicting novel and true neoantigens derived from large multi-omics data became difficult but critical challenges. The rise of Artificial Intelligence (AI) or Machine Learning (ML) in biomedicine application has brought benefits to strengthen the current computational pipeline for neoantigen prediction. ML algorithms offer powerful tools to recognize the multidimensional nature of the omics data and therefore extract the key neoantigen features enabling a successful discovery of new neoantigens. The present review aims to outline the significant technology progress of machine learning approaches, especially the newly deep learning tools and pipelines, that were recently applied in neoantigen prediction. In this review article, we summarize the current state-of-the-art tools developed to predict neoantigens. The standard workflow includes calling genetic variants in paired tumor and blood samples, and rating the binding affinity between mutated peptide, MHC (I and II) and T cell receptor (TCR), followed by characterizing the immunogenicity of tumor epitopes. More specifically, we highlight the outstanding feature extraction tools and multi-layer neural network architectures in typical ML models. It is noted that more integrated neoantigen-predicting pipelines are constructed with hybrid or combined ML algorithms instead of conventional machine learning models. In addition, the trends and challenges in further optimizing and integrating the existing pipelines are discussed.

Keywords: artificial intelligence; cancer immunotherapy; cancer neoantigen; neoantigen prediction; next generation sequencing.

Figure 1

A typical multi-layer neural network architecture showing the feature extraction and prediction procedures for neoantigen predicting. Tumor (tissue) and normal (PBMC) samples are processed by NGS-based genomic profiling (WES/WGS etc.) and undergo bioinformatic analysis to produce input data for machine learning training. The key features (tumor abundance etc.) extracted from the input data are fed to a typical neural network (deep learning) model and filters to output a predicting score (or a predicted value) for the class (positive or negative for neoantigen) the input data belongs to. The colored circles represent the input data collected from tumor components (pink, T1 to Tn) and normal cells or immune receptors (purple or sapphire, N1 to Nn), as well as the feature variables (yellow, F1 to Fn) extracted from the omics data input. The gray arrows that connect the circles shows how all the neurons are interconnected and stacked together to constitute a layer, and the multiple layers piled next to each other to construct the neural network model. Figure was created with BioRender.com.

Figure 2

A proposed neoantigen-predicting workflow implemented with machine learning (ML) models targeting individual characteristics. A group of verified neoantigen data is split into two datasets for Training + Development and Testing respectively. (A) Upper dotted line box: Model Training and Development. 1) individual features of the training data with known class (positive or negative for neoantigen) are either produced from NGS profiling directly or indirectly as additional rounds of analysis may apply to generate the variables; 2) as indicated by dashed arrow lines, these feature variables act as input in three ML models (colored boxes) targeting three characteristics: peptide-MHC binding (model a, sapphire), TCR-pMHC binding (model b, yellow) and Immunogenicity (model c, pink); 3) as indicated by the arrow lines, each model learns from its own input data and generate a prediction or together produces an integrated prediction; 4) ML model compare its prediction against the true class of the training data and learn from this training, following by optimization aiming for a better prediction. (B) Lower dotted line box: Independent Validation and Testing. After the predictive models trained and developed, a candidate neoantigen will undergo NGS-based genomic profiling and generated input data, followed by processing in the three trained models (a-c, colored clouds) and eventually provide the predictions. Figure was created with BioRender.com.