Improvement of Neoantigen Identification Through Convolution Neural Network

Area of Science:

• Immunology
• Computational Biology
• Oncology

Background:

• Accurate neoantigen prediction is crucial for developing effective cancer vaccines and T-cell therapies.
• Current prediction algorithms suffer from high false positive rates due to limitations in binding affinity data and methodology.

Purpose of the Study:

• To develop a novel deep convolutional neural network, APPM (antigen presentation prediction model), for predicting antigen presentation.
• To improve the accuracy of neoantigen identification in the context of human leukocyte antigen (HLA) class I alleles.

Main Methods:

• Developed APPM, a deep convolutional neural network model.
• Trained APPM on extensive mass spectrometry (MS) HLA-peptides datasets.
• Evaluated APPM's performance using an independent MS benchmark dataset.

Main Results:

• APPM demonstrated superior performance compared to existing methods recommended by the Immune Epitope Database (IEDB), with a higher positive predictive value (PPV) (0.40 vs. 0.22).
• Combining APPM with IEDB methods further increased the PPV to 0.51.
• Applied APPM to predict neoantigens from consensus driver mutations, identifying 16,000 putative neoantigens.

Conclusions:

• APPM significantly enhances the accuracy of neoantigen prediction.
• The model holds promise for advancing cancer vaccine and adoptive T-cell therapy development.
• Identified neoantigens possess 'driver' hallmarks, suggesting their potential role in tumor progression.