Factor analysis based methods for an integrated unsupervised analysis of multi-omics data

Multi-omics data integration is essential in life sciences for comprehensive insights into complex biological systems. With high-throughput technologies, the amount and diversity of bio-molecular data generated have grown substantially and often, multiple (omics) datasets (platforms) are available in a single experiment. To identify common biological processes across different platforms, various approaches such as network-based, cluster-based, correlation-based, and unsupervised factor-based methods have emerged in uncovering underlying patterns and interdependence within large and heterogeneous datasets. We proposed a new method and pipeline for Integrated data analysis based on Factor Analysis for Biclustering Acquisition (FABIA)[1, 2],— and conducting a benchmark analysis of this method together with other three unsupervised factor-based methods—Multi-Omics Factor Analysis (MOFA)[3], Multiple Factor Analysis (MFA)[4], Group Factor Analysis (GFA)[5]—to assess their effectiveness in integrating proteomics and other diverse bio-molecular data types. The methods were tested using both the real datasets and the data from the simulation study. Three multi-omics datasets are used: (1); an experimental dataset from the invivo radiation experiment conducted by Belgian Nuclear Research Centre (SCK•CEN) microbiology lab. to study the impact of radiation on the cognitive abilities of the newborn males mice exposed to Gy ionizing radiation. The dataset comprises of proteomics data and transcriptomics data, (2); an open-source proteomic and transcriptomics datasets, and (3); a simulated high-dimensional datasets (HDD) with pre-specified signals (latent factors). For the HDD, we examine 100 replicates, adjusting noise levels from very low to extremely high while monitoring the performance of the method's ability to identify the true signal. Our primary objective was to determine if these methods consistently identified the same patterns in terms of features and factors. The findings of our analysis revealed that all the methods uncovered similar patterns. This was assessed through pairwise-Pearson correlation analysis which examined the relationship between the feature weights and factor scores across methods. Additionally, we used Jaccard similarity indices (JSI) and performance metrics such as sensitivity, specificity, and accuracy to further evaluate the results.

References

Hochreiter, S., et al 2010. FABIA: factor analysis for bicluster acquisition.Kasim, A., et al 2016. Applied biclustering methods for big and high dimensional data using R.Argelaguet, R., et al 2020. MOFA+: a statistical framework for comprehensive integration of multi-modal single-cell data.Abdi, H., et al 2007. Multiple factor analysis (MFA). Encyclopedia of measurement and statistics.Klami, A., et al 2014. Group factor analysis. IEEE transactions on neural networks and learning systems.