How is spatial multi-omics used in drug discovery?

By measuring more than one molecular layer, such as transcript and protein abundance, on the same tissue while preserving spatial position, then checking whether the layers agree on both magnitude and location. Concordance across layers strengthens a target hypothesis, while a documented mismatch, such as a transcript signal without a matching protein signal, points to biology worth investigating rather than simply noise.

Can multi-omics improve target selection?

Yes, specifically because a single spatial layer has a structural blind spot. Spatial transcriptomics cannot reliably predict protein-level function because of variable RNA-to-protein correlation, and spatial proteomics alone offers narrower feature coverage. Combining layers on the same tissue lets a target discovery program check one layer’s signal against another before committing further resources.

What is spatial proteogenomics?

The combination of spatially resolved genomic or transcriptomic data with spatially resolved protein data from the same tissue, typically on the same section. It allows a researcher to test directly whether a gene’s expression pattern and its protein product’s abundance pattern agree in location and magnitude, rather than assuming one layer reliably predicts the other.

Spatial multi-omics for target discovery

Highlights in this article

Why multi-omics de-risks targets
The spatial omics layers
Integration approaches
Prioritizing targets with multi-omic evidence
Practical considerations

Committing to spatial multiomics rather than a single spatial layer is a decision to spend more, in tissue, time and computation, to correct a specific and well-documented blind spot: a transcript’s abundance does not reliably predict the abundance of the protein it encodes, and a target hypothesis built entirely on one layer can be wrong for a reason that layer cannot see.


Key takeaways Spatial transcriptomics offers broad feature coverage but has a structural limit predicting function, because post-transcriptional regulation makes RNA-to-protein correlation variable. Spatial proteomics measures functional molecular phenotypes more directly but with lower multiplexing capacity than transcriptomic methods. A genuine technical conflict exists between the two on the same tissue section: RNA-detection protocols often use protease treatment that damages protein epitopes needed for protein analysis. A 2024 method resolves that conflict by using spatial proteomics to guide subsequent transcriptomics capture on the same slide without compromising either signal. Graph neural network methods with dual-attention mechanisms are the current leading computational approach for combining modalities, and the number of modalities that can be co-profiled at once keeps expanding.

Key takeaways

Spatial transcriptomics offers broad feature coverage but has a structural limit predicting function, because post-transcriptional regulation makes RNA-to-protein correlation variable.
Spatial proteomics measures functional molecular phenotypes more directly but with lower multiplexing capacity than transcriptomic methods.
A genuine technical conflict exists between the two on the same tissue section: RNA-detection protocols often use protease treatment that damages protein epitopes needed for protein analysis.
A 2024 method resolves that conflict by using spatial proteomics to guide subsequent transcriptomics capture on the same slide without compromising either signal.
Graph neural network methods with dual-attention mechanisms are the current leading computational approach for combining modalities, and the number of modalities that can be co-profiled at once keeps expanding.

Why multi-omics de-risks targets

The case for combining spatial omics layers is not simply that more data is better. It rests on a specific, well-documented limitation of any single layer: none of them measures function directly, and each approximates it differently, with different blind spots.

Spatial transcriptomics offers the broadest feature coverage and genuine pathway-level insight, since it can capture expression across the whole transcriptome at once. But it faces an inherent biological limitation in predicting functional outcomes, because post-transcriptional regulation and variable RNA-to-protein correlation mean a transcript’s abundance is not a reliable stand-in for the abundance, activity or localization of the protein it encodes. A target that looks compelling by transcript abundance alone may simply reflect a gene that is transcribed heavily and translated, degraded or modified in ways the transcript never shows.

Spatial proteomics addresses that specific gap by directly capturing functional molecular phenotypes with high signal-to-noise ratios and fast data acquisition, at the cost of lower multiplexing capacity than transcriptomic methods typically achieve. Each layer, in other words, is strong exactly where the other is weak. A target hypothesis resting on only one is missing the check the other layer would have provided.

The spatial omics layers

Three spatial omics layers do most of the work in current target discovery programs, and each answers a distinct question.

Layer	What it measures	Where it is strong and where it is limited
Spatial transcriptomics	Messenger RNA abundance across the transcriptome, with tissue position preserved	Broadest feature coverage and pathway-level insight; limited by variable RNA-to-protein correlation and post-transcriptional regulation
Spatial proteomics	Protein abundance for a defined panel of targets, with tissue position preserved	Directly measures functional molecular phenotype with high signal-to-noise; typically covers fewer targets simultaneously than transcriptomic methods
Spatial metabolomics	Small-molecule and metabolite distribution across tissue	Captures downstream metabolic consequences that neither transcript nor protein abundance directly reveals; a separate mass spectrometry-based discipline with its own preparation requirements

Table 1. The three spatial omics layers most relevant to target discovery, and the specific question each answers on its own.

The metabolomics layer deserves a note on scope. It is generated by mass spectrometry imaging rather than by the sequencing- or antibody-based methods behind spatial transcriptomics and proteomics, and it carries its own distinct sample preparation, resolution and quantification considerations. Those considerations are covered in depth by our colleagues at Separation Science, who focus specifically on mass spectrometry-based spatial methods; this article treats metabolomics as one layer in an integrated dataset rather than covering its underlying technique.

Integration approaches

Combining layers on the same tissue is harder than running each separately, and the difficulty splits into two distinct problems: a technical, wet-lab problem of acquiring more than one signal from the same physical section without one measurement damaging the other, and a computational problem of reconciling the resulting datasets, which differ in resolution, dimensionality and noise characteristics.


A specific technical conflict, and how it was resolved Spatial transcriptomics protocols often use protease treatment of tissue sections to achieve efficient RNA detection, and that treatment compromises protein epitope integrity, directly threatening any protein analysis performed afterward on the same section. That is a genuine, named obstacle to combining the two layers on one slide, not a general difficulty. A 2024 method called IN-DEPTH resolves this specific conflict directly. It uses single-cell spatial proteomics to guide subsequent genome-wide spatial transcriptomics capture on the same slide, without compromising either the protein or the RNA signal, and is described as a cost-efficient and reproducible approach for this combination. The order matters: running the protein measurement first, then guiding transcriptomics capture based on it, avoids the protease exposure that would otherwise have already damaged the epitopes the protein measurement depends on.

A specific technical conflict, and how it was resolved

Spatial transcriptomics protocols often use protease treatment of tissue sections to achieve efficient RNA detection, and that treatment compromises protein epitope integrity, directly threatening any protein analysis performed afterward on the same section. That is a genuine, named obstacle to combining the two layers on one slide, not a general difficulty.

A 2024 method called IN-DEPTH resolves this specific conflict directly. It uses single-cell spatial proteomics to guide subsequent genome-wide spatial transcriptomics capture on the same slide, without compromising either the protein or the RNA signal, and is described as a cost-efficient and reproducible approach for this combination. The order matters: running the protein measurement first, then guiding transcriptomics capture based on it, avoids the protease exposure that would otherwise have already damaged the epitopes the protein measurement depends on.

On the computational side, graph-based deep learning has emerged as the leading approach for reconciling the resulting datasets. Our colleagues at Technology Networks cover this computational landscape in detail in Spatial multiomics: Methods and applications, including SpatialGlue, a 2024 method using a graph neural network with a dual-attention mechanism that integrates spatial location and omics measurements within each modality before combining modalities through cross-omics integration. Benchmarked across epigenome-transcriptome and transcriptome-proteome datasets from different tissue types, SpatialGlue more accurately resolved spatial domains, including cortical layers in the brain and macrophage subsets in the spleen, compared with analysis of either modality alone. The number of modalities that can be co-profiled simultaneously continues to expand quickly: a 2025 method demonstrated five-modality co-profiling from a single tissue section, covering two histone modifications, chromatin accessibility, the whole transcriptome and a targeted protein panel at once.

Prioritizing targets with multi-omic evidence

Once transcriptomic, proteomic and, where available, metabolomic layers are integrated on the same tissue, prioritization becomes a question of concordance and discordance rather than a single-layer expression threshold.

Concordant signals across layers strengthen confidence. A candidate elevated in transcript abundance, protein abundance, and downstream metabolic activity, all in the same tissue location, is a substantially stronger hypothesis than any one signal alone.
Discordant signals are informative, not just noise to average away. A transcript elevated without a corresponding protein signal may point to active post-transcriptional regulation worth understanding mechanistically, rather than simply disqualifying the candidate outright.
Spatial domain agreement matters as much as expression level agreement. Methods such as SpatialGlue are explicitly built to resolve whether different modalities agree on where a tissue domain begins and ends, not only whether they agree on expression magnitude within an already-assumed domain.

That discordance point is worth dwelling on, since it runs against the instinct to treat disagreement between layers as simply a data quality problem. A transcript-protein mismatch driven by genuine post-transcriptional regulation is exactly the kind of biology a single-layer analysis would never surface, because a single layer has no second signal to disagree with.

Practical considerations

Four practical constraints shape whether a spatial multi-omics approach is the right investment for a given target discovery question.

Cost and tissue consumption scale with layer count. Each additional spatial layer adds acquisition cost, and same-slide approaches consume tissue that a single-layer study would not need, which matters when starting material, such as a rare clinical biopsy, is limited.
Resolution and platform compatibility differ by layer. Not every spatial transcriptomics platform is compatible with every spatial proteomics approach at matched resolution, and confirming platform compatibility before committing to a same-slide design avoids a costly mismatch discovered after the fact.
Computational expertise is a real prerequisite, not an afterthought. Graph-based integration methods require computational skills many wet-lab-focused target discovery teams do not have in-house, which argues for building or accessing that capability before generating multi-omic data rather than after.
Not every target question needs every layer. A question specifically about whether a transcript-level signal reflects genuine functional biology benefits from a matched protein layer; a question already validated at the protein level may not need an additional transcriptomic pass to justify the added cost.

The tissue-context foundation this spoke builds on, including how a single spatial layer already improves on bulk and dissociated approaches, is covered in Finding drug targets in the tissue context with spatial biology. For the broader argument connecting tissue context to target validation and clinical success evidence, see Spatial biology for target discovery: Mapping disease at tissue resolution, and for where target discovery sits within the full spatial biology pipeline, Spatial biology in drug discovery: From target discovery to translational medicine.

This article was produced under Drug Discovery News’s editorial policies.