CMB_2024v14n2

Computational Molecular Biology 2024, Vol.14, No.2, 64-75 http://bioscipublisher.com/index.php/cmb 66 3 Methods for Multi-Omics Data Integration 3.1 Statistical methods Statistical methods for multi-omics data integration are foundational and often serve as the first step in understanding complex biological systems. These methods include various forms of regression analysis, Bayesian statistics, and factor analysis. For instance, Bayesian models can incorporate prior knowledge and handle different data distributions, making them suitable for integrating heterogeneous omics data (Li et al., 2016). Additionally, methods like Multiple Canonical Correlation Analysis (MCCA) and Multiple Factor Analysis (MFA) are used to identify correlations and shared variations across different omics layers, facilitating a more comprehensive understanding of biological interactions (Tini et al., 2019). 3.2 Machine learning approaches Machine learning (ML) approaches have gained significant traction in multi-omics data integration due to their ability to handle large, complex datasets and uncover hidden patterns. These approaches can be broadly categorized into supervised learning, unsupervised learning, and deep learning techniques. Supervised learning techniques are employed when the outcome variable is known, and the goal is to predict this outcome based on input features from multiple omics datasets. Common methods include support vector machines (SVMs), random forests, and various forms of regression analysis. For example, SVMs and random forests have been effectively used to integrate genomic, proteomic, and metabolomic data for disease prediction and biomarker discovery (Feldner-Busztin et al., 2023). These methods are particularly useful in precision medicine, where they can help in patient stratification and personalized treatment plans (Reel et al., 2021). Unsupervised learning techniques are used when the outcome variable is unknown, and the goal is to uncover the underlying structure of the data. Clustering methods like k-means, hierarchical clustering, and more advanced techniques like Similarity Network Fusion (SNF) are commonly used. SNF, for instance, integrates multiple types of omics data by constructing similarity networks for each data type and then fusing them into a single network, which can reveal complex biological relationships (Figure 1) (Tini et al., 2019). Other methods like Multiple Co-Inertia Analysis (MCIA) and Joint and Individual Variation Explained (JIVE) are also employed to identify shared and unique variations across different omics datasets (Tini et al., 2019). Deep learning methods, particularly neural networks, have shown great promise in multi-omics data integration due to their ability to model complex, non-linear relationships. Autoencoders, a type of neural network, are frequently used to transform multi-omics data into latent representations that capture essential features while reducing dimensionality (Hauptmann and Kramer, 2022). For example, methods like MOLI, Super.FELT, and OmiEmbed have been developed to integrate multi-omics data for drug response prediction, showing that deep learning can outperform traditional methods in certain contexts (Hauptmann and Kramer, 2022). Additionally, novel architectures like Omics Stacking combine the advantages of intermediate and late integration, further enhancing predictive performance (Figure 2) (Hauptmann and Kramer, 2022). Customizable deep learning strategies, such as CustOmics, adapt training to each data source independently before learning cross-modality interactions, providing interpretable results and high performance in tasks like tumor classification and survival outcome prediction (Benkirane et al., 2023). 3.3 Network-based methods Network-based methods are another powerful approach for integrating multi-omics data. These methods construct networks where nodes represent biological entities (e.g., genes, proteins) and edges represent interactions or associations between them. Network-based diffusion/propagation methods can exploit information captured in each omics dataset to infer associations between different data types (Cominetti et al., 2023). For instance, network-based models have been used to integrate genomic, transcriptomic, and proteomic data to identify key regulatory pathways and potential therapeutic targets in cancer (Nicora et al., 2020). These methods can also incorporate external knowledge from biological databases, enhancing the robustness and interpretability of the results (Vahabi and Michailidis, 2022).

RkJQdWJsaXNoZXIy MjQ4ODYzNA==