In response to the significant structural differences between heterogeneous data types and the difficulty of traditional integration methods in extracting interpretable features within a unified framework, this study proposes an improved computational biology oriented multi-omics fusion algorithm. By constructing a multi-layer heterogeneous network consisting of the genome, transcriptome, and proteome, and sequentially introducing deep random walk feature modeling, an improved iterative DBSCAN density clustering algorithm, and a cross layer relationship prediction model, a unified representation and association inference of multi-omics networks are achieved. Using public datasets of rice Magnaporthe oryzae interactions, we generated co-expression links, miRNA-mRNA regulatory links, and protein interactions. The deep random walk model captures 92% of local hierarchical structures within four steps, outperforming Node2Vec by 14%. The improved DBSCAN achieved 28% and 35% gains in Silhouette and Calinski-Harabasz scores, respectively. Cross-layer link prediction reached an AUC of 0.87, exceeding SNF (0.76) and iCluster+ (0.70). GO and KEGG analyses confirmed the biological validity of predicted key factors, with enriched pathways related to protein ubiquitination and endoplasmic reticulum stress response. This method provides an interpretable and reproducible computational framework for heterogeneous multi-omics integration, with broad applicability to complex biological interaction studies.
本文所提“深度随机游走”(Deep random walk,DRW)与经典DeepWalk算法[18]在概念上存在本质区别:经典DeepWalk的“深度”源于其借鉴深度学习中的Skip-gram模型进行节点嵌入,而本文DRW的“深度”则特指3个层面——网络结构的深度,即游走运行于包含基因、miRNA、蛋白质的多层异质网络之上,需穿透不同组学层以捕获层内拓扑与跨层语义双重信息,而非局限于单一同构网络;游走策略的深度,即区别于Node2Vec仅通过超参数控制游走倾向的做法,DRW引入层类型权重矩阵 Φ,使转移概率能根据生物学先验知识对不同层间跳转施加差异化偏好(如强化miRNA→mRNA的跨物种调控连接),引导游走路经聚焦于生物意义明确的深层关联模式;特征建模的深度,即DRW生成的路径序列不仅编码节点ID,更融合节点特征(如表达量)与边权(如互作置信度),使得后续学习到的嵌入向量能反映多层异质网络中的深层拓扑与语义特征。以下详述DRW的算法设计与执行流程。在完成多层异质网络构建的基础上,本文设计了一个3阶段的分析流程(图2),以解决异质网络整合的核心挑战:跨层特征对齐与高维稀疏关系推断。首先,通过深度随机游走(DRW)提取并对齐不同组学层的节点特征;其次,利用卷积特征增强的DBSCAN聚类对高维特征进行去噪与功能模块识别[19,20];最后,构建一个融合多源信息的评分函数,实现跨层生物关系的精准预测[21]。
深度随机游走模块。转移概率计算中,层类型权重矩阵 Φ 的查表操作为常数时间O(l),边权Wv,u为预计算值,因此单步游走的采样时间为O(1),单条游走序列生成时间为O(l)。全部r×n条序列的总生成时间为O(r×l×n)。Skip-gram嵌入训练采用负采样策略,设负采样数为k,每个节点-上下文对的计算复杂度为O(d×k),总训练复杂度为O(r×l×n×d×k)。该模块的空间复杂度由嵌入矩阵 Z ∈ℝn×d主导,为O(n×d)。
JiangHai-ping, GaoChun-chun, LiuWen-hao, et al. Progress in life science research driven by data[J]. Bulletin of the Chinese Academy of Sciences, 2024, 39(5): 862-871.
HuaXia. Process intensification and regulation mechanism of whole-cell catalytic oxidation of biomass alcohols/aldehydes[D]. Nanjing: College of Chemical Engineering, Nanjing Forestry University, 2024.
ShiYao, DuanMao-teng, LiuBing-qiang, et al. Application of graph transformer models in multi-omics data integration analysis[J]. Mathematical Modeling and Its Applications, 2025, 14(2): 1-8.
GengJie, HouChuan-dong, ChenHao-ran, et al. Construction of miRNA-mRNA regulatory network and prediction of targeted therapeutic drugs for hemophilia A[J]. Journal of Zhengzhou University (Medical Science Edition), 2025, 60(6): 754-759.
LinHuan, CaoMei-hua. Identifying key genes in hepatocellular carcinoma by weighted gene co-expression network analysis combined with protein-protein interaction analysis[J]. Journal of Medical Theory and Practice, 2025, 38(22): 3794-3800.
ZhouWei, LiuZhi-cheng. Integration technology of iCluster migration imaging subsystem[C]∥Proceedings of the 26th Annual Meeting of the Chinese Geophysical Society & the 13th Academic Annual Meeting of the Chinese Seismological Society, Ningbo, China, 2010: 408.
ZhangLi-li. Research on cancer survival analysis model based on graph neural network[D]. Xi'an: School of Science, Xi 'an University of Architecture and Technology, 2025.
Zhi-yaoLa, QianYu-rong, LengHong-yong, et al. Overview of research on graph embedding based on random walk[J]. Computer Engineering and Applications, 2022(13): 1-13.
LuoShao-hui, LuoYi-jun. Research on engineering vehicle trajectory clustering based on KMeans++ and DBSCAN algorithm[J]. Urban Geotechnical Investigation & Surveying, 2023(2): 27-30.
ShengYi-tao, ZhengXiao-xuan. Research on base station site selection and sector angle planning based on DBSCAN clustering[J]. Journal of Nantong Vocational University, 2022, 36(4): 64-68.
WanXiao-mei. Physiological and ecological response characteristics and adaptation mechanisms of alfalfa to simulated calcium habitats[D]. Guizhou: College of Geography and Environmental Sciences, Guizhou Normal University, 2025.
WangYe-jian, LiangXiao-ling, AbulaAbulaiti, et al. Identification of drought-responsive miRNAs and their target genes during tassel floral organ differentiation in maize by deep sequencing[J]. Xinjiang Agricultural Sciences, 2020, 57(8): 1373-1384.
WuEn-hui, QiaoLiang. Microbial metaproteomics: from sample preparation and data acquisition to data analysis[J]. Chinese Journal of Chromatography, 2024, 42(7): 658-668.
ShengJian-hua. Research on key biomolecular recognition algorithms for rice resistance to Magnaporthe oryzae[D]. Changchun: College of Computer Science and Technology, Jilin University, 2024.
ZhangTian-yue. Research on key factor network construction algorithm for Magnaporthe oryzae-rice interaction based on multi-omics[D]. Changchun: College of Computer Science and Technology, Jilin University, 2023.
HanCui-yan, HaoYue, LiDe-yong, et al. Road engineering coordination for multivariate spatio-temporal object clustering analysis[J]. Geospatial Information, 2024, 22(11): 49-52.
[39]
ZhangH, ZhaoE, LiL, et al. GERWR: identifying the key pathogenicity-associated sRNAs of Magnaporthe oryzae infection in rice based on graph embedding and random walk with restart[J]. IEEE/ACM Transactions on Computational Biology and Bioinformatics, 2024, 21(2): 227-239.
[40]
CrandallS G, FilgueirasC, GoldK M, et al. A multi-omics approach to solving problems in plant disease ecology[J]. PLoS One, 2020, 15(8): e0237975.
[41]
Cembrowska-LechD, KrzeminskaA, MillerT, et al. An integrated multi-omics and artificial intelligence framework for advanced plant phenotyping in horticulture[J]. Biology, 2023, 12(10): 12101298.
[42]
NijsM, De MoorB, WaelkensE. Non-negative matrix factorization for the analysis of mass spectrometry imaging data (with applications to other omics modalities)[R]. Lirias Report LIRIAS4159628, Leuven: KU Leuven, 2024.
JiKai-song, MaHong, WangGeng-run, et al. Stay point extraction from user trajectory based on bi-directional long short-term memory network[J]. Journal of Information Engineering University, 2021, 22(6): 641-646.