全基因组互作效应对揭示复杂性状的遗传基础至关重要，然而，在当前用于解析和预测复杂性状、疾病的基因组统计模型中，丰富的基因互作信息常被忽略。为弥补这一不足，本研究提出了一种名为biBLUP（基于生物学互作效应的最佳线性无偏预测）的新型上位效应模型。该模型通过重点关注KEGG通路内的基因间互作，从而整合了先验的生物学知识。模拟实验证明，在多种遗传结构下，biBLUP模型能有效捕获互作效应，与忽略生物学互作信息的传统模型相比，其预测准确性最高可提升62%。本研究利用跨物种的真实数据进一步验证了biBLUP模型的性能。例如，在对6,642个酵母株系数据的分析中，biBLUP模型通过模拟与尿囊素利用相关的KEGG通路内的遗传互作效应，使生长速率的预测准确性提高了40.36%。此外，在水稻开花期的研究中，将KEGG信息整合到biBLUP模型中，不仅成功捕获了已验证的上位效应，还使开花期的预测准确性提升了16.29%。本研究结果表明，将KEGG通路信息整合到基因组预测模型中，能够有效捕获具有生物学意义的互作效应，从而提高模型的预测能力，并加深对复杂性状遗传基础的理解。本研究的核心创新在于提出了biBLUP新模型，将KEGG通路这一生物学先验知识整合进基因组预测模型中，专门用于模拟基因间的互作效应。与传统模型不同，biBLUP模型通过聚焦特定生物通路内的互作，精准捕获了具有生物学意义的遗传互作效应。经模拟和跨物种真实数据验证，该方法不仅显著提升了复杂性状的预测准确性，还成功解析了已知的上位性机制，为深入理解复杂性状的遗传基础提供了有力的新思路。

Although genome-wide interaction effects are critical for unraveling the underlying genetic architectures of complex traits, the rich landscape of biological interactions is often disregarded in statistical models for genomic dissecting and predicting complex traits/diseases. To bridge this gap, we introduce biBLUP (biological interaction Best Linear Unbiased Prediction), a novel epistatic model that integrates prior biological knowledge by focusing on interactions among genes within KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways. Simulation experiments demonstrate that biBLUP effectively captures interaction effects across diverse genetic architectures, achieving up to a 62% increase in predictive accuracy compared to models ignoring such information. We validated the performance of biBLUP using real data across species. In a specific application using data from 6,642 yeast lines, biBLUP yielded a 40.36% improvement in prediction accuracy for growth rate by modeling genetic interaction effects within the KEGG pathway associated with allantoin utilization. Furthermore, incorporating KEGG into biBLUP successfully captures validated epistatic effects associated with rice flowering time. This integration results in an improvement of 16.29% in prediction accuracy for flowering time of rice. Our findings demonstrate that integrating KEGG pathway information into genomic prediction models enables the capture of biologically relevant interaction effects, thereby enhancing both predictive ability and our understanding of the genetic basis of complex traits.