DeepImmuno Tunes into CNN to Predict T-Cell Immune Response
Published November 2021 | Briefings in Bioinformatics
The DeepImmuno artificial intelligence platform developed by experts at Cincinnati Children’s provides superior results when employing the convolutional neural network (CNN) model to predict how T-cells will recognize constantly mutating cancer cells and ever-evolving invasive pathogens. A study led by Guangyuan Li, PhD, Nathan Salomonis, PhD, and colleagues, benchmarked DeepImmuno’s performance using five traditional machine learning models (ElasticNet, K-nearest neighbors, support vector machine, Random Forest and AdaBoost) and three deep learning models (CNN, Residual Net and graph neural network).
They found that the DeepImmuno-CNN model was the best at simulating how non-native peptides would interact with specific immune system genes that activate T-cell responses. Rather than attempting to sift through all potential interactions between peptides and major histocompatibility complex (MHC) genes, this tool predicts the immunogenicity of MHC-peptide pairs.
“In addition to outperforming two highly used immunogenicity prediction algorithms, DeepImmuno-CNN correctly predicts which residues are most important for T-cell antigen recognition and predicts novel impacts of SARS-CoV-2 variants,” says Li, corresponding author for the study.
The published work updates information about DeepImmuno that had been shared as a preprint in 2020 during the early days of the COVID-19 pandemic. This tool is useful for researchers seeking to stay a step ahead of emerging mutations of the SARS-CoV-2 virus. It also can assist scientists working to develop cancer immunotherapies, which requires accurately predicting which cancer-specific neopeptides are most likely to elicit an immune response. An artificial intelligence tool is needed for this work because thousands of potential disease-associated antigens can be presented in innate or foreign cells. Tools like DeepImmuno-CNN can help prioritize which candidates are most likely to induce a T-cell response prior to experimental validation.
