Multi-omic integration analyses measurements from different molecular layers, such as DNA, RNA and protein, side by side while keeping track of what each one measured and which specimen it came from.
A DNA change, an RNA count and a protein or phosphorylation measurement answer different questions, so integration starts with an evidence record for each measurement: the patient or experimental unit, specimen, collection time, tissue region, method, units and quality limits. Measurements from one specimen are paired; measurements from separate cohorts are not, and even adjacent tissue sections do not hold identical cells. Three reports that copy one DNA call add no new layer, and a value predicted by a model, such as RNA imputed from an image, must stay labelled as imputed beside measured values. Multi-Omics Factor Analysis (Argelaguet and colleagues, 2018) is one research method: it infers hidden factors that capture biological and technical sources of variability, separates axes shared across data types from axes specific to one, and can impute missing data; the authors applied it to 200 samples of chronic lymphocytic leukaemia profiled for somatic mutations, RNA expression, DNA methylation and ex vivo drug response. A factor is a statistical pattern, not a mechanism or a validated test. Technical variation needs its own handling: Risso and colleagues (2014) showed that usual RNA sequencing normalisation mostly accounts for sequencing depth and fails to correct library preparation and other unwanted technical effects, and where treatment time and processing method always change together, those data alone cannot separate their effects. In a proteogenomic study of 105 genomically annotated breast cancers, 77 of which gave high-quality data, pathway analysis of the phosphoproteome identified a G-protein-coupled receptor cluster that was not readily identified at the mRNA level (Mertins and colleagues, 2016), an example of layers adding different information. Correlation between layers is not a tested dependency, and a controlled intervention addresses a different question. Layers need not agree numerically, and disagreement is not automatically an error: bulk RNA from a mixed sample can be high for a gene whose protein, on a tissue image, sits mainly in normal vessel cells, so averaging the two into one confidence score would hide whether malignant cells carry an accessible target. Combining layers can reveal agreement, competing explanations, missing evidence or a sample mismatch, and it does not by itself show that a matching treatment will help: the ESMO Scale for Clinical Actionability of molecular Targets (2018) grades alterations in six levels of clinical evidence by their implications for patient management, from targets ready for routine decisions to investigational targets that need more data (Mateo and colleagues, 2018; a second version appeared in 2026). For a prediction model, a patient who appears in several layers must stay on one side of a patient-level split, and many cells from one biopsy are not many patients.
Showing the technology this term belongs to: Proteomics & phosphoproteomics.
Shares Deep mass-spectrometry proteome (CPTAC), mRNA to protein concordance, Proteomics & phosphoproteomics.
Shares Batch effects and harmonisation, Bulk RNA sequencing (RNA-seq) and the full transcriptome.
Shares Batch effects and harmonisation, Data leakage in model evaluation.
Shares Tumour purity, Spatial transcriptomics.
Shares Multimodal fusion (early, late, modality dropout), Data leakage in model evaluation.
Shares Batch effects and harmonisation, Bulk RNA sequencing (RNA-seq) and the full transcriptome.
Shares Tumour purity, Cell composition confound (tumour versus stroma and immune cells).
Shares Deep mass-spectrometry proteome (CPTAC), mRNA to protein concordance, Proteomics & phosphoproteomics.
Every source this page cites, grouped by who published it. Grouping is read from the web address, not assigned by hand.
A source that has moved or gone is reported by the weekly link check. /audit/