Synthetic Engineering with ImmunoGenAI: Multi-Modal Deep Learning Integration for Predicting and Optimizing Immune Evasi

Overview

The human leukocyte antigen (HLA) system comprises over 20,000 alleles, making immune recognition one of the most polymorphic biological processes. Therapeutic proteins encode overlapping peptide fragments processed by major histocompatibility complex (MHC) molecules, where allele-specific binding shapes immune risk. Although derived from a therapeutic sequence, minor amino acid substitutions can substantially alter binding landscapes, increasing probability of immune activation. Today, many personalized therapeutics–including monoclonal antibodies, engineered enzymes, and cancer vaccines–experience reduced efficacy from immune-mediated clearance, neutralizing antibodies, or inflammation. Synthetic engineering of immune evasion is paramount for improving biologics. However, industry-standard frameworks are predominantly prediction-only systems that evaluate peptide-HLA binding solely, or optimization pipelines that apply local sequence edits without utilizing sequential and contextual information. Such approaches can reduce scores while disrupting structural-translational stability. In this research, a novel multimodal deep learning integration developed by fusing six modality-specific encoders trained on peptide-HLA immunopeptidomics. Over 100,000 nonredundant, experimentally valid peptide-HLA entries and structural chains were used for deep learning. The transformer multimodal utilizes multi-head self-attention to evaluate sequence dependencies and immune interactions. ImmunoGenAI is evaluated on 20 vaccine antigens and a benchmark set of at least length 53 therapeutic proteins previously characterized. ImmunoGenAI’s performance across immunogenicity, epitope density, structural preservation, optimization efficiency, and population coverage is compared to that of five industry techniques. ImmunoGenAI achieved a 43.1% increase in immune evasion across therapeutic sequences. This research demonstrates that multimodal deep learning prediction and optimization reduces immune evasion while maintaining stability, establishing a feasible and scalable personalized therapeutic device.

Competition history

• CSEF 2026 Medicine & Physiology (Track 2) (Senior Division) · Entry S-21-03

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Biomedical and Health Sciences projects
• CSEF Biomedical and Health Sciences projects
• Every CSEF 2026 project