Peptide Discovery from Spatial and Single-Cell Data

Treating Cell Maps as a Prioritization Layer, Not a Gene List

The most useful shift is mental: regard the omics output as a prioritization layer that selects where biology concentrates, and reserve the peptide-design decisions for after that selection is validated. Building peptide discovery from spatial and single-cell data rather than from bulk differential-expression lists is what lets a program target a real cell state instead of an average. In drug discovery practice, single-cell RNA sequencing has become a standard layer for nominating and prioritizing targets because it resolves variation that bulk sequencing averages away. Reviews of applications of single-cell RNA sequencing in drug discovery describe how scRNA-seq supports target identification, cell-subtype-specific analysis, и, critically, early attrition risk by showing which cells respond and how dosing affects them. Spatial methods then add the context bulk data destroy: preserving high-resolution spatial information is what lets analysts see new cell types, cell interactions, and tissue structures that only make sense once a cell’s location is kept.

For a peptide target, the actionable layers collapse to a short list:

At the end of this stage you hold a prioritized candidate list, not a verdict. The verdict comes from the validation layers that follow.

Prioritizing a Peptide Target: Where Cell-State Specificity Earns Its Keep

A target that is merely differentially expressed across the whole sample is a weak starting point for a peptide, because a peptide’s value usually lives in its selectivity. That intuition has quantitative support. A large retrospective analysis of single-cell data across human tissues found that cell-type specificity and disease-cell specificity are both features enriched among targets entering clinical development, and that cell-type specificity in the disease-relevant tissue was robustly predictive of target progression from Phase I to Phase II. The result is emerging and preprint-level, and it deserves measured phrasing — but it aligns with a practical truth: a receptor that is transcriptionally restricted to the cell state you care about, and to the place that state lives, is the one most likely to give a peptide a clean on-target window.

Prioritization therefore should reward three properties at once:

• State specificity. The receptor should be enriched in the disease-relevant cell state relative to neighboring and phenotypically similar states, not merely present in the tissue.

• Spatial restriction. It should concentrate in a niche that is physically reachable by the intended route, which simultaneously reduces the population of off-target cells a ligand will encounter.

• Surfaceome membership. Because peptides act on extracellular or accessible targets, prioritize genes whose products are known membrane proteins rather than intracellular or secreted species.

A useful filter is to score each candidate on these three axes and advance only the shortlist that clears all of them. A receptor that is state-specific but buried, or surface-exposed but ubiquitous, is a poor first bet no matter how impressive its log-fold change looks.

Cell-State Specificity: The Protein-Level Reality Check

Here is where many programs quietly tip over. Transcriptomic cell-state specificity is a hypothesis about protein reality; it is not proof of it. A gene can be strongly up in a cell state and still not be present as the displayable, on-the-surface protein that a peptide needs to bind. Choosing a cell-state-specific peptide target therefore requires proving that the transcript-level specificity survives translation and trafficking — not assuming it does.

The literature on cell-selective ligand engineering quantifies why this matters. Cell-population selectivity is best defined functionally — as the ratio of ligand bound to the target-cell population over ligand bound to off-target populations — and achieving it depends on the target’s actual surface presentation, not its transcript count. And because peptides generally must reach an extracellular epitope, target accessibility can matter as much as expression level; a highly expressed protein buried in a complex or trapped intracellularly is out of reach.

The protein-level validation stack, in practice, is short and orthogonal:

• Flow cytometry on intact, non-permeabilized cells is the first-line check that the receptor sits on the exterior surface and quantifies cell-to-cell heterogeneity. High-throughput versions are well established for cell-surface profiling.

• Immunofluorescence or immunohistochemistry confirms membrane localization and distinguishes a genuine surface signal from an intracellular pool.

• Surface proteomics or targeted mass spectrometry gives unbiased protein-level evidence and can surface isoforms or candidates that lack good antibodies.

• Proteo-genomic methods such as CITE-seq-style readouts pair each cell’s RNA with its surface protein, and single-cell proteo-genomic reference maps exist precisely to validate that an RNA-defined cell state carries the predicted surface protein.

One timing note: a cellular thermal shift assay (CETSA) is often mentioned alongside these tools, but it detects ligand-induced protein stabilization — it is a target-engagement readout that assumes a binder already exists. Use it after you have a candidate, not as a surface-exposure gate.

Key Takeaway: Treat differential expression as the motivation to look, then prove the protein is on the surface of the intended cell state before you design a single peptide around it. Cell-state specificity that survives a protein-level check is the specificity a screen can actually detect.

Selecting a Ligand Axis: Confirming the Receptor–Ligand Pair

Once a surface-accessible, state-specific receptor survives validation, the design question becomes which axis to engage — and here the temptation is to trust an inference tool’s output as if it were a binding event. It is not.

Tools such as CellPhoneDB и CellChat curate ligand–receptor interaction databases and infer which communication axes are statistically enriched between cell groups in single-cell or spatial data. They are excellent for prioritizing which sender–receiver pairs are worth the effort of testing, and CellChat in particular models multisubunit complexes, agonists, antagonists, and co-receptors. But an inferred axis is a candidate, not a confirmed interaction. Panning a peptide against a receptor and seeing binding does not by itself prove the peptide is a functional cell-targeting ligand, and a computationally predicted pair needs orthogonal confirmation before it earns synthesis budget.

The confirmation package for a ligand axis is:

• Competition or displacement evidence, or another orthogonal binding readout, that the peptide and the presumed receptor genuinely interact.

• A counter-screen panel built early, spanning closely related receptor subtypes and relevant off-target cells. Affinity is not selectivity; subtype selectivity is established only by testing the candidate against the wider receptor family, as design work on increased receptor-subtype selectivity has made explicit.

• State-aware design for conformationally dynamic receptors. Many peptide targets such as GPCRs exist in active and inactive conformations, and ligand preference can depend on which state is populated. Designing against the desired state rather than a static structure can be the difference between an agonist that drives the biology you want and one that does not.

The compact decision test is: is the receptor enriched in the target state, is it surface-accessible on native cells, and is there direct evidence the peptide binds it more strongly than close homologs and off-target tissues? A “no” on any of these is a reason to redesign the axis before committing to material.

The Pre-Synthesis Evidence Package: A Go/No-Go Gate

By this point the program has a validated, accessible receptor and a confirmed ligand axis. The discipline most teams really need is a single, explicit gate that must clear before an expensive synthesis and screening spend is authorized.

The pre-synthesis evidence package consolidates the checks above into an auditable list. It is deliberately short, because an evidence gate that is too heavy gets skipped, and one that is too light lets weak candidates through:

A candidate that clears every row is worth synthesizing and screening. One that fails even a single row should be redesigned, re-derived, or dropped rather than forced through. The costly failures in peptide discovery are rarely sequences that were obviously bad — they are sequences that looked perfectly good on one axis and were never tested on the others until after the synthesis spend.

From Candidate to Material: Readiness for Synthesis and Screening

The final judgment is about the material itself, and it is worth separating chemical readiness from biological readiness so the two do not get conflated.

Chemical readiness is a matter of defined quality thresholds, and the standards differ by stage. For discovery screening, a common research acceptance bar is chromatographic purity of at least 95% by reversed-phase HPLC or UHPLC, with roughly 98–99% reserved for clinical-grade material; peptide identity and purity are normally established by orthogonal reversed-phase HPLC-UV and mass spectrometry rather than a single assay. Because purity standards and impurity expectations differ so sharply between a research screen and a regulated program, the analytical peptide testing and QC plan should be set before, not after, the first delivery.

Biological readiness is where the cell-map work pays off. A peptide that reaches screening should carry the confirmation it earned upstream: the surface-accessible state-specific receptor, the confirmed interaction, and a first-pass Kd and selectivity figure. Without those, a positive screen signal is ambiguous and a negative one is uninformative. That is why the highest-value habit in this workflow is to gate synthesis behind the evidence package rather than ahead of it.

This is also the point where a candidate that was prioritized from maps meets the realities of the custom peptide synthesis bench and comprehensive peptide services — sequence design, modification strategy, solubility-aware analog design, and batch-level HPLC/MS documentation all become decision inputs that determine whether the material your assay sees is the molecule your biology intended.

Pro Tip: Order a small initial panel of close analogs — not a single sequence — so your first synthesis gives you an SAR signal, a solubility hedge, and a selectivity readout in one round instead of several.

Building Peptide Discovery from Spatial and Single-Cell Data Around the Evidence, Not the Enthusiasm

What separates a peptide discovery program that converts high-resolution biology into moving candidates from one that collects interesting maps is a refusal to conflate signal with solution. The maps nominate; validation selects; and an explicit evidence gate decides when a candidate is worth expensive material.

The order of operations matters more than any single tool. Prioritize a surface-accessible, state-specific receptor. Prove it is on the surface of the intended cell type. Confirm the ligand axis you plan to engage. Assemble a short auditable package — relevance, surface presence, confirmed interaction, Kd, selectivity, and developability — and only then spend on synthesis and screening.

When your panel of candidates clears that gate, the next consideration is whether your synthesis and QC partner can deliver the orthogonally characterized, batch-documented material a screening program needs — the kind of discipline that keeps the assay measuring your molecule and not the noise. That is where a partner such as MOL Changes offering end-to-end custom peptide synthesis with Class 100 sterile manufacturing, peptide testing and QC, and a broad catalogue of modifications can help — so the peptide your screen sees is the one your map promised. Aligning the biology decision with the chemistry execution from day one is what turns a cell map into a candidate program that actually moves.