CRISPR-ImmunoPred: A Multi-Stage AI Framework to Predict T-Cell Reactivity and Mitigate Immunogenicity in CRISPR-Cas9 Therapies

The Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR) associated protein 9 (Cas9) system has revolutionized and changed the genome editing by enabling precise and efficient genetic modifications across various diverse biological applications. Despite its ability, the broader therapeutic adoption of CRISPR-Cas9, particularly in vivo settings, is constrained drastically by immunogenicity risks arising from guide RNA (gRNA) target interactions and unintended host immune recognition patterns. Accurate prediction of these T-cell epitope reactivity is essential to ensure safety, mitigate adverse immune responses in application, and guide the rational design of genome-editing therapies. In this work, we propose CRISPR-ImmunoPred, a phased hybrid prediction framework that integrates gradient boosting and deep learning for immunogenicity-aware CRISPR design framework. The framework comprises of (i) a Particle Swarm Optimization (PSO) optimized XGBoost based on gRNA Viability Profiler for assessing CRISPR editing fitness and (ii) an Attentive Immunogenicity Inference model which employs cross-attention Transformer architectures to explicitly capture guide target sequence interactions alongside numerical biological features. Comprehensive sensitivity analysis, systematic ablation studies, and group-wise generalization experiments finally demonstrate that each architectural component contributes meaningfully to predictive performance and that the proposed framework degrades gracefully under biologically meaningful distributional shifts. On combining high predictive accuracy with partial interpretability through feature attribution and architectural analysis, CRISPR-ImmunoPred provides a robust and modular approach for balancing editing efficiency and immune safety, supporting safer CRISPR-based therapeutic development and rational genome editing design.