What Is Single-Cell RNA Sequencing? A Complete Guide for Researchers

Inquiry

Summary

Single-cell RNA sequencing (scRNA-seq) measures gene-expression programs in individual cells rather than averaging signals across an entire specimen. It can reveal cell types, transient states, response programs, developmental paths, and rare populations that are difficult to resolve with bulk RNA sequencing. This guide explains what the method measures, how a project moves from sample to interpretation, which design choices shape the data, and when single-nucleus or spatial approaches may answer the research question more directly.

The central planning principle is simple: begin with the biological contrast and the unit of inference, then work backward to sample handling, replication, capture strategy, sequencing, and analysis. A large cell count cannot compensate for weak biological replication, poorly preserved material, or a comparison confounded with processing batch. The purpose of scRNA-seq is not to produce the largest possible cell atlas; it is to generate interpretable evidence at the resolution the study requires.

Figure 1. A single-cell RNA sequencing project connects tissue handling and cell or nucleus preparation with molecular capture, a cell-by-gene expression matrix, and biological interpretation of cell populations and states.

Key Takeaways

• scRNA-seq resolves mixtures. It assigns expression profiles to individual captured cells, allowing common and rare populations to be examined separately.
• The sample determines the strategy. Freshness, tissue architecture, cell fragility, storage history, and expected cell types influence whether whole cells or nuclei are more appropriate.
• Replication remains biological. Thousands of cells from one donor are many observations from one specimen, not thousands of independent biological replicates.
• Quality control is multi-stage. Sample condition, suspension quality, library complexity, cell-level metrics, ambient RNA, doublets, and batch structure all affect interpretation.
• Method choice follows the question. Bulk RNA-seq, scRNA-seq, snRNA-seq, full-length single-cell RNA sequencing, and spatial transcriptomics answer related but different questions.

Single-Cell RNA Sequencing at a Glance

In a typical scRNA-seq experiment, intact cells are separated, transcripts from each captured cell receive a cell-specific barcode during library preparation, and sequencing reads are summarized as a matrix of genes by cells. Computational analysis then groups cells by shared expression patterns, annotates populations, tests condition-associated changes, and reconstructs selected biological processes. Modern workflows differ in capture scale, transcript coverage, sensitivity, and compatibility with additional molecular measurements, but they share this core logic [1].

The word "single-cell" describes the resolution of the molecular profile, not necessarily perfect isolation of every biological cell. A captured droplet may contain two cells, damaged cells can release RNA into the surrounding solution, and closely related populations may remain difficult to separate. The observed dataset is therefore a filtered and sampled representation of the original specimen. Good studies make that sampling process visible through prespecified quality checks and careful reporting rather than treating the final matrix as a literal census.

scRNA-seq is most useful when the research question depends on heterogeneity. Examples include distinguishing immune and stromal responses within a perturbed tissue, identifying a rare progenitor population, comparing the same cell type across conditions, or testing whether an apparent bulk expression shift reflects cellular composition rather than regulation within cells. Researchers who need a broader service-level view can review the single-cell sequencing service portfolio, while projects centered on droplet-based expression profiling can examine the 10x Genomics Chromium single-cell RNA-seq workflow.

What Can scRNA-Seq Reveal That Bulk RNA-Seq Can Miss?

Bulk RNA-seq reports the average transcript abundance across all cells contributing RNA. That average is valuable for well-defined specimens and can offer deep, cost-efficient expression profiling, but it combines changes in cell abundance with changes occurring inside a cell type. If a marker rises in bulk tissue, the cause could be more marker-positive cells, stronger expression per cell, or both. scRNA-seq separates these possibilities by assigning expression patterns to individual captured cells before aggregation.

This resolution makes hidden mixtures visible. Two cell states with opposing responses can cancel each other in bulk data; a rare state can be diluted below detection; and a modest shift in tissue composition can resemble a regulatory response. At single-cell resolution, researchers can compare matched cell populations across samples and ask whether the same biological process is shared or confined to a subset. The benefit is not simply "more detail." It is the ability to frame the statistical comparison around a defined cellular context.

Figure 2. Bulk RNA-seq provides an averaged expression profile, scRNA-seq separates cellular profiles, and spatial transcriptomics adds tissue location. The appropriate resolution depends on the biological question rather than a universal hierarchy of methods.

Single-cell resolution also supports discovery of continua rather than only discrete types. Activation, differentiation, stress, and cell-cycle programs can vary gradually. Dimensionality-reduction plots are useful views of those patterns, but distances on a plot are not physical distances, and cluster boundaries depend on analytical choices. Claims should be confirmed through marker coherence, consistency across specimens, alternative parameter settings, and, when possible, orthogonal measurements [2].

The choice among averaged, single-cell, and spatial measurements is developed further in the resource on bulk, single-cell, and spatial transcriptomics. In many studies, these methods are complementary: bulk RNA-seq can provide efficient cohort-level screening, scRNA-seq can define cell-resolved programs, and spatial profiling can test where those programs occur in intact tissue.

How Does Single-Cell RNA Sequencing Work?

A project begins before library preparation. Tissue collection, transport time, temperature, dissociation chemistry, wash conditions, and sorting can all alter which cells survive and which transcripts are detected. A clean suspension should contain mostly intact, individual cells at a concentration compatible with the selected capture method. Samples are then partitioned so that cellular transcripts can be associated with a cell barcode, converted into sequencing libraries, and read on a sequencing instrument.

After sequencing, software associates reads with genes and cell barcodes to construct the expression matrix. Empty partitions and low-information barcodes are removed; likely doublets, damaged cells, and heavily contaminated profiles are evaluated; and accepted cells are normalized for downstream comparison. The workflow then moves from technical processing to biological analysis: feature selection, dimensionality reduction, neighborhood construction, clustering, annotation, differential testing, pathway analysis, and selected advanced models. The exact sequence should follow the question, not a fixed menu [3].

Three aspects of the measurement deserve special attention:

• Capture is incomplete. Not every transcript present in a cell is observed, especially for low-abundance genes. A zero can mean no expression or missed detection.
• Cells are sampled, not exhaustively measured. Recovery depends on tissue dissociation, cell size, fragility, filtering, and loading. Some populations may be depleted before sequencing begins.
• Expression is a snapshot. The profile reflects the state at collection and processing. It does not by itself establish a temporal sequence, protein abundance, functional activity, or causal regulation.

These constraints do not negate the method; they define what conclusions are proportionate to the evidence. A well-designed project uses controls, biological replication, metadata, and validation to distinguish technical artifacts from reproducible biology. Detailed validation options are discussed in approaches for validating scRNA-seq findings.

What Samples Can Be Used for scRNA-Seq?

Fresh tissue, cultured cells, organoids, sorted populations, and some cryopreserved cell suspensions can be compatible with whole-cell scRNA-seq when viable single cells can be recovered. Blood and dissociable cultures often present fewer physical barriers than fibrotic, adipose-rich, calcified, or densely connected tissues. Yet compatibility cannot be inferred from tissue name alone. Collection medium, ischemic interval, storage, dissociation history, cell concentration, debris, viability, and the biological populations of interest all matter.

Archived frozen tissue usually cannot be returned to an intact viable-cell suspension, but nuclei can often be released for single-nucleus RNA sequencing (snRNA-seq). Nuclei are also useful for tissues in which dissociation selectively damages large, fragile, multinucleated, or highly connected cells. The resulting expression profile is enriched for nuclear and pre-messenger RNA and may have lower representation of some cytoplasmic transcripts, so scRNA-seq and snRNA-seq should not be treated as technically identical. Researchers working with frozen or difficult tissue can review single-nucleus RNA sequencing services.

No sample specification should be accepted as a guarantee of usable data. A small pilot can be more informative than a theoretical compatibility statement, particularly for rare material or unfamiliar species.

scRNA-Seq Experimental Design: Decisions to Make Before the Sample Is Processed

The first decision is the unit of biological replication. If the question concerns treatment response across animals, donors, cultures, or experimental batches, those independent specimens are the replicates. Individual cells nested within one specimen do not replace them. Statistical tests that ignore this nesting can produce overly confident results because cells from the same specimen share biology and handling history. Donor-aware aggregation or mixed models are often more appropriate for condition-level inference, and false discoveries can rise when cell-level tests are used as if every cell were independent [6].

Balance is equally important. Condition, collection day, operator, reagent lot, and processing lane should not align in a way that makes the experimental contrast indistinguishable from batch. Multiplexing strategies can distribute specimens across common processing units, but sample identity, expected composition, and demultiplexing performance need to be considered. When perfect randomization is impossible, the limitation should be recorded before analysis rather than discovered after clustering.

Figure 3. Project design moves from the biological contrast and replicate structure to sample compatibility, capture scale, sequencing allocation, analysis, and validation. Decisions made upstream constrain every downstream comparison.

Before processing, the team should agree on the following:

• Primary comparison. Define the cell population, condition, time point, and outcome that will support the main conclusion.
• Biological replicates. Specify independent specimens per group and distinguish them from technical repeats or multiple captures of one sample.
• Expected heterogeneity. Estimate whether rare populations, continuous states, or dominant cell types drive the required recovery.
• Sample route. Choose whole cells, nuclei, or an alternative based on storage history and tissue behavior.
• Batch plan. Randomize or balance conditions across collection, preparation, library construction, and sequencing.
• Validation plan. Decide which key cell populations, markers, or regulatory claims will be checked with an independent method.

Power cannot be summarized by a single recommended number of cells. Detecting a population that represents one percent of recovered cells has different sampling requirements from comparing a major cell type across donors. Differential-expression power depends on replicate number, effect size, within-group variability, cell abundance, read allocation, and the analysis model. Experimental-design tutorials emphasize aligning these components with the intended inference rather than using a platform-wide default [8].

Cell recovery and read allocation also need to be planned together. Loading more cells spreads a fixed sequencing budget across more profiles, which can improve sampling of population frequencies while reducing information available for each cell. A project aimed at identifying broad cell classes may tolerate shallower profiles than one centered on subtle state differences or low-abundance transcripts. The correct balance depends on the expected expression signal, sample number, and whether the key analysis compares abundance, expression, or both. Pilot saturation curves and downsampling can help determine whether added reads or added biological specimens are more likely to improve the main endpoint.

Covariates should be collected before the analysis plan is finalized. Age, sex, genotype, collection site, time point, tissue region, preservation interval, and processing metadata can explain real variability or expose confounding. Not every recorded variable belongs in every statistical model, especially in a small cohort, but missing metadata cannot be reconstructed from the expression matrix. A concise data dictionary linking specimen, preparation, library, and sequencing identifiers prevents common errors when samples are pooled or reprocessed.

Key Quality Control Checkpoints

Quality control begins with the specimen and continues through interpretation. At the suspension stage, microscopy or an automated counter can assess concentration, aggregates, debris, cell integrity, and apparent viability. Those observations should be interpreted together: a high viability estimate does not rescue a suspension dominated by clumps, and an apparently clean suspension may still have lost fragile populations. For nuclei, membrane removal, nuclear integrity, clumping, and tissue-derived debris require separate attention.

Library-level checks evaluate whether the expected fragment distribution and sequencing-ready material were obtained. After matrix construction, commonly reviewed cell-level features include detected genes, total transcript counts, mitochondrial transcript fraction for whole cells, and patterns consistent with doublets or ambient RNA. Universal cutoffs are risky because expected values differ by tissue, species, chemistry, cell type, and preparation. Thresholds should be justified from the observed distributions and biological context, then tested for their effect on cell composition [2].

A useful QC report keeps specimen identity visible. Pooling metrics across all cells can hide a failed replicate, and removing an outlying sample after seeing the biological result can introduce bias. Quality criteria should be applied consistently, with any exception documented. For outsourced analysis, researchers should expect sample-level summaries, filtering rationale, and traceable outputs rather than only a final embedding.

Quality is also population-dependent. A threshold that removes cells with relatively few detected genes may disproportionately exclude quiescent lymphocytes, while a high-count cutoff may remove legitimate large or transcriptionally active cells together with doublets. Mitochondrial transcript fractions can reflect damage, normal tissue biology, or both. Sensitivity analysis should therefore compare annotations and condition effects under reasonable alternative filters. If a finding appears only after one narrow threshold choice, it should be treated as fragile until independently confirmed.

Ambient RNA deserves separate evaluation because it can create low-level expression of abundant tissue markers in unrelated cells. The pattern is often recognizable when a highly expressed gene appears diffusely across many clusters without its accompanying program. Statistical correction can reduce estimated contamination, but it should be supported by empty-partition profiles, tissue knowledge, and before-and-after views. Correction is least reliable when the presumed contaminant is also genuinely expressed at low levels in the receiving population.

What Does scRNA-Seq Data Analysis Usually Include?

The analysis begins with read processing and creation of a gene-by-cell matrix, but most biological conclusions arise later. Quality filtering removes profiles unlikely to represent intact single cells. Normalization and feature selection support comparisons across a wide dynamic range. Dimensionality reduction and graph-based clustering organize similar profiles, while annotation combines established markers, reference datasets, and context-specific knowledge. Best-practice reviews stress that no single step is neutral: parameter choices, reference quality, and batch correction can all change the apparent structure [4] [7].

Figure 4. Common outputs include quality summaries, cell maps, annotations, marker patterns, donor-aware comparisons, pathways, trajectories, and interaction hypotheses. Each output answers a different question and carries different assumptions.

A well-scoped analysis may include:

• Quality assessment and filtering with sample-level metric distributions and documented thresholds.
• Cell-type and state annotation supported by multiple markers and relevant references rather than one gene alone.
• Differential abundance to test whether the representation of a defined population changes across replicated conditions.
• Differential expression within a cell type using models that preserve the experimental unit.
• Pathway or gene-set analysis to summarize coordinated expression changes without treating every enriched label as a mechanism.
• Trajectory or velocity analysis when the sampling design and biology plausibly represent a dynamic process.
• Cell-cell communication inference as a hypothesis-generating analysis based on expressed ligand-receptor pairs, not direct proof of signaling.

Integration across specimens can reduce technical variation and support shared annotation, but overcorrection may remove real condition-specific biology. Analysts should compare integrated and unintegrated views, preserve sample metadata, and test whether major findings recur across replicates. The single-cell RNA-seq data analysis service can support projects requiring a documented pipeline, sample-aware comparisons, and interpretable deliverables.

Annotation should be considered an evidence synthesis task. Automated label transfer is useful when a reference matches the species, tissue, developmental stage, and assay, yet reference labels may be broader than the new dataset or may omit a condition-specific state. Manual marker review adds biological context but can become subjective. A defensible workflow reports the reference, similarity confidence, marker combinations, unresolved populations, and any labels assigned at different levels of granularity. Sometimes "stromal cell" is better supported than a more specific subtype, and preserving that uncertainty is preferable to overannotation.

Downstream models should match the output they claim to produce. A trajectory orders cells along an expression manifold; it does not directly observe lineage unless supported by time, labeling, or lineage tracing. Communication analysis identifies compatible expression of signaling partners; it does not measure secretion, receptor binding, or response. Regulatory-network analysis links factors with candidate targets using statistical structure and prior knowledge. These tools are valuable when presented as ranked hypotheses with assumptions, not as automatic mechanistic proof.

Major Research Applications

Cell atlases are a visible application of scRNA-seq, but the method is equally useful in focused hypothesis-driven studies. Developmental research can resolve transitional populations and lineage-associated expression programs. Immunology studies can separate activation states within phenotypically related cells. Tumor and microenvironment research can examine malignant, immune, stromal, and vascular compartments without averaging their signals. Drug research can identify which cell populations respond, fail to respond, or express a target-associated program, provided that expression is not mistaken for protein activity or efficacy.

Cross-species and non-model-organism research has also expanded as genome annotation and computational resources improve [5]. Here, orthology, annotation completeness, tissue sampling, and species-specific cell composition become part of the design. A cell type with a familiar label may not have an identical expression program across species, so translation should be tested at the level of conserved and divergent programs rather than assumed from naming alone.

Common research objectives include:

• mapping cell composition and states in tissues, organoids, and experimental systems;
• resolving treatment-associated responses within defined populations;
• identifying candidate markers for enrichment, imaging, or functional follow-up;
• testing how genetic perturbations alter specific cellular compartments;
• linking transcriptional states with chromatin, protein, immune-receptor, or spatial measurements;
• constructing reference datasets for comparative or longitudinal studies.

These applications remain observational unless the study includes perturbation, temporal evidence, or functional validation capable of testing causality. A transcription factor whose targets are enriched in a cell state is a candidate regulator, not automatically the driver of that state. Likewise, an inferred ligand-receptor pair identifies a plausible communication route but does not demonstrate physical interaction. Careful wording improves both scientific credibility and the usefulness of the result for follow-up experiments.

Longitudinal questions require special design because destructive sequencing does not follow the same individual cell over time. Repeated tissue collections usually sample related but different cells, and differences can reflect composition as well as temporal regulation. Time points should be biologically justified and processed in a balanced way. Trajectory methods can connect intermediate states within the sampled data, but the resulting path is a model of transcriptional similarity. A time course, perturbation, or lineage measurement is needed to determine whether cells actually move along that path.

Multi-omic extensions can connect expression with surface proteins, immune receptors, chromatin accessibility, or spatial position. Their value depends on whether the added layer tests a specific uncertainty. For example, paired chromatin and RNA data can evaluate whether a state-associated expression program coincides with altered regulatory accessibility, while spatial data can test whether a population localizes to a predicted niche. Integration is strongest when sample pairing and measurement resolution are explicit; a correlation between separate cohorts is weaker than a link measured in the same cells or matched sections.

Common Failure Modes and How to Reduce Them

The most damaging failures often occur upstream of sequencing. Selective cell loss can make an abundant tissue population appear rare. Extended dissociation can induce stress-response transcripts that are then misread as biology. Cell aggregates increase multiplets, while damaged cells release RNA that blurs population-specific expression. These risks are reduced through tissue-specific optimization, short and consistent handling, temperature control appropriate to the protocol, gentle cleanup, and pilot assessment of representative material.

Design failures can be less visible. Processing all control specimens on one day and all treated specimens on another creates a confounded comparison. Combining cells from several donors without preserving donor identity prevents valid replicate-level inference. Capturing many cells from too few specimens produces an impressive matrix but weak evidence for population-level claims. These problems cannot be fully repaired computationally after data collection.

Analytical failure modes include aggressive filtering that removes a real fragile population, circular annotation based only on expected markers, clustering at a resolution selected to match a desired narrative, and differential testing that treats cells as independent replicates. Risk reduction depends on transparent sensitivity analyses:

• compare cell composition before and after filtering;
• inspect marker combinations rather than isolated genes;
• test whether conclusions persist across reasonable clustering resolutions;
• retain specimen identity throughout all plots and models;
• distinguish exploratory cell-level findings from donor-level confirmatory tests;
• validate high-priority observations with an independent assay or dataset.

Published guidance on experimental design and analysis provides useful starting points [4], but local decisions still require biological judgment. The strongest QC threshold is not the strictest one; it is the threshold that removes technical failures while preserving interpretable biology and behaves consistently across samples.

scRNA-Seq, snRNA-Seq, Full-Length RNA, or Spatial Transcriptomics: How to Choose

No assay is universally superior. Droplet-based scRNA-seq supports broad profiling of many dissociated cells and is often selected for cellular composition and state analysis. snRNA-seq can recover profiles from frozen or hard-to-dissociate material and may reduce some dissociation-associated biases, although nuclear RNA coverage changes the detectable transcriptome. Full-length single-cell protocols can provide richer transcript structure for fewer cells, which may matter for isoform or allele-focused questions. Spatial transcriptomics preserves location but may trade cellular isolation, transcript coverage, field of view, or target breadth depending on the platform.

Choice becomes clearer when the question is written as a sentence: "We need to compare X in Y population across Z independent specimens." If Y cannot be recovered as intact cells, nuclei may be preferable. If the conclusion depends on location, dissociation-based data alone are incomplete. If the main outcome is a cohort-level average in a purified population, bulk RNA-seq may be more efficient. For projects combining dissociated and spatial profiles, see the guide to integrating scRNA-seq with spatial transcriptomics.

When scRNA-Seq May Not Be the Best First Choice

scRNA-seq may be unnecessary when the specimen is already homogeneous and the question concerns average expression across many replicates. It may also be premature when collection procedures are unstable, specimen metadata are incomplete, or the tissue cannot yield a representative cell suspension. In those situations, a pilot, bulk assay, nuclei-based approach, targeted panel, or spatial method may answer the immediate question with fewer assumptions.

The method is also not a direct measurement of protein abundance, chromatin state, metabolite concentration, cell morphology, or tissue position. Expression can nominate mechanisms and targets, but additional modalities are needed when the conclusion rests on those molecular layers. Multi-omic designs should be driven by a specific integration question; collecting every available modality without a predefined link between them can increase complexity without improving inference.

Practical reasons to pause before choosing scRNA-seq include:

• the primary contrast has too few independent biological specimens;
• condition and processing batch cannot be separated;
• the target population is unlikely to survive dissociation and nuclei are not informative for the required genes;
• tissue location is essential to the hypothesis;
• the expected effect concerns isoforms or variants not captured well by the selected workflow;
• there is no feasible plan to validate the main biological claim.

A method-selection discussion should surface these limitations early. Changing the assay before processing is far easier than explaining after sequencing why the data cannot support the intended conclusion.

Planning a Single-Cell RNA Sequencing Project With CD Genomics

An effective project consultation starts with the biological comparison, specimen inventory, collection history, expected cell populations, and required outputs. From there, the study can be mapped to whole-cell or nucleus preparation, an appropriate capture strategy, sequencing allocation, and a donor-aware analysis plan. The proposal should state assumptions and decision points, especially when sample quality or rare-population recovery cannot be known in advance.

CD Genomics supports research workflows spanning sample preparation, single-cell and single-nucleus sequencing, and bioinformatics. Service planning can include feasibility review, pilot design, sample-level quality assessment, cell annotation, replicated condition comparisons, and integration with spatial data. Deliverables and analysis depth should be specified before processing so that sequencing and computation are matched to the decisions the study needs to make.

The most useful information to prepare includes:

• tissue or cell type, species, preservation method, and collection timeline;
• number of independent specimens and the planned biological contrasts;
• available cell counts or tissue amounts and any prior dissociation results;
• cell populations or genes that are central to the hypothesis;
• need for whole cells, nuclei, spatial context, or additional molecular modalities;
• required outputs, validation strategy, and publication or data-sharing expectations.

Data delivery should also be planned at the start. Researchers may need raw sequencing files, processed count matrices, sample and cell metadata, quality-control tables, analysis notebooks or parameter records, figures, and editable result tables. File naming and metadata keys should remain consistent across these outputs so that a cell barcode can be traced back to its specimen and processing unit. Reproducibility is easier when reference genome and annotation versions, software versions, random seeds, filtering rules, and model formulas are recorded as the analysis is performed.

Finally, the project should define what would count as an interpretable negative result. Failure to find a condition effect may reflect a truly small effect, insufficient biological replication, loss of the responsive population, weak detection of the relevant genes, or excessive variability. Prespecified QC and power assumptions help distinguish these possibilities. A negative result with adequate sampling and a detectable positive-control program can be informative; a negative result from an unrepresentative suspension usually is not.

This planning framework is offered for research use. It does not constitute clinical testing, diagnosis, treatment selection, or patient-specific interpretation.

Frequently Asked Questions

References

1. Cole AG. Establishing single cell RNA transcriptomics: a brief guide. Frontiers in Zoology. 2025;22(1):25. doi:10.1186/s12983-025-00579-x.
2. Kim GD, Lim C, Park J. A practical handbook on single-cell RNA sequencing data quality control and downstream analysis. Molecules and Cells. 2024;47(9):100103. doi:10.1016/j.mocell.2024.100103.
3. Lim J, Park C, Kim M, et al. Advances in single-cell omics and multiomics for high-resolution molecular profiling. Experimental & Molecular Medicine. 2024;56(3):515-526. doi:10.1038/s12276-024-01186-2.
4. Heumos L, Schaar AC, Lance C, et al. Best practices for single-cell analysis across modalities. Nature Reviews Genetics. 2023;24(8):550-572. doi:10.1038/s41576-023-00586-w.
5. Woo H, Eyun SI. Applications and techniques of single-cell RNA sequencing across diverse species. Briefings in Bioinformatics. 2025;26(4):bbaf354. doi:10.1093/bib/bbaf354.
6. Squair JW, Gautier M, Kathe C, et al. Confronting false discoveries in single-cell differential expression. Nature Communications. 2021;12(1):5692. doi:10.1038/s41467-021-25960-2.
7. Luecken MD, Theis FJ. Current best practices in single-cell RNA-seq analysis: a tutorial. Molecular Systems Biology. 2019;15(6):e8746. doi:10.15252/msb.20188746.
8. Lafzi A, Moutinho C, Picelli S, Heyn H. Tutorial: guidelines for the experimental design of single-cell RNA sequencing studies. Nature Protocols. 2018;13(12):2742-2757. doi:10.1038/s41596-018-0073-y.

Research Use and Trust Statement

This article is intended for research use only. It does not provide medical advice and is not intended for diagnostic, prognostic, preventive, or therapeutic use. Method suitability, sample feasibility, and analytical scope should be evaluated for each research project, and literature-derived associations should not be interpreted as proof of causality without appropriate validation.