Estimating genotype-tissue specific gene expression using hybrid deep learning.

Genotype-tissue expression profiles are critical for understanding how genetic variation influences gene regulation across tissues, yet they are often missing or incomplete, and experimental profiling is costly and time-consuming. Although computational approaches exist for multi-tissue imputation and sequence-based expression prediction, they do not explicitly use expression information from neighboring reference genes and their genomic context for collated multi-tissue imputation. To address this, we developed a novel hybrid deep learning model that integrates a convolutional neural network (CNN), a transformer encoder, and an XGBoost regressor to estimate these profiles with high accuracy. By combining promoter sequences, tissue correlations, intergene distances, and gene orientation, our model achieves ~30% higher accuracy than distance-based methods, generating expression profiles that closely align with experimental data. We demonstrate its utility by completing missing profiles in the GTEx dataset. Our model offers a practical and scalable alternative to experimental profiling and enables cost-effective estimation of genotype-tissue-specific expression profiles, particularly for lowly expressed RNA genes and less-characterized genomes, paving the way for advances in genomics research where experimental data are scarce.