NeoaPred: a deep-learning framework for predicting immunogenic neoantigen based on surface and structural features of...

Area of Science:

• Computational biology
• Immunoinformatics
• Cancer research

Background:

• Neoantigens are crucial for cancer immunotherapy, arising from somatic mutations and presented by human leukocyte antigen (HLA) to T cells.
• Current bioinformatic methods for identifying immunogenic neoantigens lack satisfactory accuracy.
• Peptide-HLA class I (pHLA-I) complex surface and structural features provide key insights into neoantigen immunogenicity.

Purpose of the Study:

• To develop an accurate deep-learning framework for predicting immunogenic neoantigens.
• To leverage structural and surface features of pHLA-I complexes for improved neoantigen identification.
• To enhance the development of effective cancer immunotherapies.

Main Methods:

• Developed NeoaPred, a deep-learning framework for predicting neoantigens.
• Accurately constructed pHLA-I complex structures, with 82.37% achieving RMSD < 1 Å.
• Integrated surface, structural, and atom group features of mutant vs. wild-type peptides to calculate a foreignness score.

Main Results:

• NeoaPred achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.81 and an Area Under the Precision-Recall Curve (AUPRC) of 0.54 on the test set.
• The foreignness score effectively predicted neoantigen immunogenicity.
• Outperformed existing neoantigen prediction methods.

Conclusions:

• NeoaPred offers a significant advancement in neoantigen prediction accuracy.
• The framework's ability to model pHLA-I complexes and integrate feature differences is key to its performance.
• NeoaPred has the potential to accelerate the development of personalized cancer immunotherapies.