No Universal Winner: Running Seven Multi-Omics Integrators Head-to-Head on BioMate
Newsroom · BioMate Showcase

No Universal Winner: Running Seven Multi-Omics Integrators Head-to-Head on BioMate

GenomicsAug 17, 20267 min readBioMate Team

If you remember one thing about multi-omics integration, make it this: there is no universal winner. The two flagship benchmarks — Rappoport & Shamir 2018 and Cantini 2021 — both conclude the best method depends on the dataset and the question. Our Multi-Omics Integration survey maps the full landscape; the practical consequence is what BioMate is built for.

Why "pick one method" is the wrong move

Factor models (MOFA, iCluster) give interpretable, view-attributable factors; network fusion (SNF) is nonlinear and noise-robust; non-negative factorization (intNMF) is best at recovering ground-truth clusters; MCIA is the most consistent all-rounder. Because the leader changes by cohort, the defensible workflow is to run several and compare them on your own data — exactly the thing that's tedious to wire up by hand.

What BioMate does

  • Seven SOTA integrators, one interface. MOFA2, SNF, intNMF, MCIA, MFA, iClusterPlus, and JIVE run on AWS Batch (Nextflow), each invoked through a universal CSV interface (features × samples per modality, sample-intersected across views).
  • Head-to-head, scored against ground truth. A worked pilot compared them on real chronic lymphocytic leukemia (matched to IGHV status) and TCGA (matched to PAM50): MOFA2 led on CLL, MCIA led on TCGA — the leader changes by cohort, exactly as the benchmarks predict.
  • SOTA normalization, automatically. Per-modality normalization (log2-CPM for bulk RNA, log1p-CP10k for scRNA, M-value for methylation, TF-IDF for scATAC, median-centering for proteomics) is applied before integration — fixing the failure mode where one high-magnitude view swamps the rest.
Universal CSV per modality→SOTA normalization→7 integrators in parallel→Metric harness vs labels→Best method + subtypes
Figure. How BioMate compares integrators — one CSV interface runs seven methods and scores them against ground truth.
The bottom line. Instead of betting on one integrator, BioMate runs seven and shows which one actually works on your data, scored against ground truth — the comparison the benchmarks say you need, delivered as a one-command job with a wet-lab loop behind it.

References

  1. Rappoport N, Shamir R. Multi-omic and multi-view clustering: review and cancer benchmark. Nucleic Acids Res 2018;46:10546. doi:10.1093/nar/gky889
  2. Cantini L, et al. Benchmarking joint multi-omics dimensionality-reduction approaches. Nat Commun 2021;12:124. doi:10.1038/s41467-020-20430-7
  3. Wang B, et al. Similarity network fusion for aggregating data types on a genomic scale. Nat Methods 2014;11:333. doi:10.1038/nmeth.2810
  4. Argelaguet R, et al. Multi-Omics Factor Analysis (MOFA). Mol Syst Biol 2018;14:e8124. doi:10.15252/msb.20178124

Go deeper: the full survey

This showcase is one half of a pair. For the complete, citation-backed map of the field — every method, its strengths and limits, and what is actually winning — read the Multi-Omics Data Integration survey.

Try BioMate free
Read the survey: Multi-Omics Data Integration → · All surveys: Research Surveys → · More: Newsroom