A few days ago, I got a call from a friend asking if I can help her plan some multi-omic analysis (I believe the exact quote was: “don’t you do multi-omics something something?”). She had a high-level goal of better understanding a biological system, and to that end she was planning to obtain six (yes, six, you are reading correctly) different omic profiles for her samples. So she knew what her overall goal was, and she was quite certain about the types of data she needed, but when it came to the actual data analysis and integration, she was completely in the dark.
From conversations with colleagues and collaborators in human microbiome research over the years, this seems to be a common situation: researchers have the end goal in mind and the data in hand, but struggle to identify the relevant methods or analysis approaches to use. You can also see this reflected in many microbiome multi-omic papers, where each omic is analysed separately, and the “integration” ends up being more manual and interpretation-based.
In the Borenstein lab, this is something we discussed often. As a group that develops computational methods for multi-omic integration, we followed the literature on such methods, and all agreed it has become quite overwhelming: new methods are published frequently, and the landscape is so branched that it’s nearly impossible to follow. Some excellent review papers in recent years have helped map this space, categorizing methods based on their statistical or algorithmic framework, or on the specific omics being analysed. As we thought about it more, we realized there are actually multiple ways to categorize multi-omic integration methods, and, just to make things more complicated, these “axes” are largely orthogonal. Methods can be grouped by their statistical approach, by data characteristics (number and type of omics), by the resolution of integration (feature-level vs. global patterns), or by the stage at which integration happens (do we first analyse each omic independently and then combine results? or do we first combine them and then analyse jointly?).
Nonetheless, we felt that the main barrier is earlier than that: getting from a high-level objective (like my friend had) to concrete research questions that then allow you to clearly formulate the problem. For the most part (with some exceptions), multi-omic methods can actually be organized around the types of questions they are designed to answer. We thought that if researchers had a clearer map of these question types, and the methods that best address each question, that could be a much more practical starting point.
So with that goal in mind, we wrote this review. Word limits meant we couldn’t’ include many exciting methods out there (the original version had nearly 100 additional references…. but tough decisions had to be made!), but we hope it can serve as a useful guide for researchers who feel overwhelmed by the many approaches labelled “multi-omic integration”, just as we did.
(Image credit: ChatGPT)