Getting more from multimodal omics
Getting more from multimodal omics
Unlocking the full potential of multimodal omics to get the highest return on investment.
Have you ever felt like you're not quite getting to the meaningful biological discoveries and tangible conclusions you expected to get from your multimodal omics? It's easy to get lost in the complexity and just stick to old recipes. We end up missing out on proper discovery of MoA, novel targets or combination strategies, and biomarkers for patient selection. Here is an alternative view of the discovery process, showcasing how we can unlock the full potential of multimodal omics and get the highest return on our investment.
I've been diving deep into the world of single-cell genomics over the past few years, employing a diverse arsenal of analytical and experimental strategies to enhance our understanding of anti-tumor immunity in patients. I identified novel molecular programs, decoded T cell clonality patterns, and uncovered how the spatial positioning and interaction patterns of rare immune cell subsets may drive response or resistance to immunotherapy (Magen et al. Nat Med (2023)). This has led to some interesting implications for future treatments which are now being explored in a number of contexts across academia and industry. But more often than not, the analysis and interpretation process is highly inefficient, and the outcomes are not so informative.
One of the biggest issues I've noticed is the way the bioinformatic modeling process is sometimes perceived by biologists as a "black box". It's like some magical machine that spits out unpredictable results influenced by mysterious parameters. Even in the dry lab, we can fall into the trap of using 'one-size-fits-all' templated workflows that we find in tutorials or various software applications. These workflows give a false sense of simplicity, but they often lead to poorly interpretable or even misleading results.
Peeling the layers of noise and bias
The truth is, extracting meaningful insights from multimodal omics is akin to peeling an onion. The valuable signal is obscured by layers of bias and noise which are unique to each biological system and experimental setup. There is no magic bullet; it's a meticulous process of separating the signal from these unwanted influences, layer by layer, until we strike gold. While some issues have to be mitigated through improved experimental design, the rest have to be managed with a more nuanced approach to data analysis and interpretation.
Take for example the common challenge of clustering cells based on their transcriptomic profiles in the single-cell genomics realm (scRNAseq and other modalities). This approach frequently fails to identify meaningful cell types and states due to various confounders that extend beyond the well-known cell cycle effects. Likewise, the results of differential expression in human scRNAseq data are often dominated by noise and bias. This varies depending on the biological and experimental context, again emphasizing the need for a context-specific solution rather than a universal process.
The three pillars for effective discovery
So, what's the alternative? The approach that transformed my ability to extract meaningful conclusions from multimodal omics revolves around three pillars: effective signal decomposition, multimodal harmonization, and experimental design for inference. These pillars form a cyclical process, and enhancing the efficiency of this iteration cycle is your key to reaching meaningful conclusions, and doing so in the fastest way possible.
Over the next few weeks, I'll delve into each of these pillars and share my perspectives. I encourage researchers from diverse biological and computational backgrounds to join the conversation. If your discovery efforts could benefit from my help, don't hesitate to reach out.
Stay tuned.
ASSAF MAGEN, PhD
Building something radically different to unlock the power of multimodal omics.
Founder & CEO, Voyant Bio