# Quality Control Metrics for Single-Cell Proteomics Mass Spectrometry: What to Check Before Analysis Single-cell proteomics by mass spectrometry generates complex datasets that require rigorous quality assessment before any biological interpretation. This article defines the essential quality control metrics that researchers should evaluate prior to downstream analysis, including protein identification rates per cell, quantitative reproducibility, empty droplet rates, and data completeness. The practical outcome is a structured framework for deciding whether a single-cell proteomics dataset is suitable for further processing or whether experimental or computational adjustments are needed. ## Scope and Reader Context Researchers working with single-cell proteomics data face a distinct problem: mass spectrometry instruments produce raw files that contain signals from individual cells, but distinguishing genuine single-cell measurements from background noise, contamination, and technical artifacts requires systematic quality evaluation. The decisions made at this stage determine whether downstream analyses such as clustering, differential expression, and pathway interpretation produce reliable biological conclusions. This article serves biology students, researchers, laboratory professionals, and life-science practitioners who need concrete metrics and thresholds for assessing data quality before investing time in complex analytical workflows. The focus is on practical quality control metrics that can be computed from standard mass spectrometry output files, with attention to the limitations of each metric and the circumstances under which professional escalation is warranted. The scope covers data inputs from typical single-cell proteomics experiments, including those generated by nano-proteomic sample preparation methods and multiplexed data acquisition strategies. The quality control principles described here apply across different instrument platforms and experimental designs, although specific threshold values may require adjustment based on the particular method used. ## Core Principles of Single-Cell Proteomics Quality Control ### Why Quality Control Differs from Bulk Proteomics Bulk proteomics experiments analyze millions of cells simultaneously, which averages out technical variation and provides abundant material for protein identification. Single-cell proteomics operates with vastly smaller amounts of starting material, often containing only picogram quantities of total protein per cell. This fundamental difference creates quality control challenges that do not exist in bulk analyses. The limited material available from a single cell means that many proteins fall below the detection limit of the mass spectrometer. A typical single-cell proteomics experiment may identify only a fraction of the proteins present in the cell, and the subset of proteins detected can vary substantially between cells. This variability is both biological and technical in origin, and distinguishing between these sources requires careful quality assessment. The [standardized workflow using the scp package](https://pubmed.ncbi.nlm.nih.gov/38907155) emphasizes that data analysis approaches differ substantially between research teams and that evaluating pipelines is difficult because ground truth datasets are often missing. The developers of this workflow addressed this challenge by creating a design with cell lines mixed in known proportions to generate controlled variability for benchmarking purposes. This approach highlights the importance of having reference datasets with known composition when developing and validating quality control procedures. ### The Role of Data Structures in Quality Assessment Quality control in single-cell proteomics depends on the data structures used to organize and process the measurements. The [scp package](https://pubmed.ncbi.nlm.nih.gov/38907155) relies on the QFeatures and SingleCellExperiment data structures, which are widely used in the Bioconductor ecosystem. These structures allow researchers to track measurements at different levels of processing, from raw features through peptides to proteins, and to associate quality metrics with individual cells. The [Bioconductor project](https://bioconductor.org/) provides official documentation for these data structures and the packages that support them. Understanding how data are organized within these frameworks is essential for implementing quality control procedures correctly, because each processing level requires different quality metrics and thresholds. ### Quality Control as a Multistage Process Quality assessment in single-cell proteomics is not a single step but rather a process that occurs at multiple stages of data processing. The [scp workflow](https://pubmed.ncbi.nlm.nih.gov/38907155) describes quality control at both the feature level and the cell level, followed by aggregation of raw data into peptides and proteins, normalization, and batch correction. Each stage has specific quality metrics that inform decisions about whether to proceed, filter, or repeat measurements. The feature-level quality control examines individual mass spectrometry signals before they are assigned to peptides and proteins. The cell-level quality control evaluates the overall performance of each single-cell measurement, including how many proteins were identified and how consistent the measurements are across technical replicates. This multistage approach ensures that problems are detected at the earliest possible point in the workflow. ## At a Glance: Essential Quality Control Metrics The following table summarizes the primary quality control metrics that should be evaluated before downstream analysis of single-cell proteomics data. These metrics provide a practical starting point for assessing dataset quality, with the understanding that specific thresholds may vary depending on the experimental method and instrument platform. | Quality Metric | What It Measures | Typical Assessment Approach | Action When Out of Range | | --- | --- | --- | --- | | Proteins per cell | Depth of proteome coverage achieved for each individual cell | Count identified proteins after filtering for confidence and reproducibility | Low counts may indicate sample loss, insufficient injection, or acquisition problems | | Coefficient of variation | Quantitative reproducibility across technical replicates or similar cells | Compute standard deviation divided by mean for protein quantities | High variation suggests inconsistent sample preparation or instrument performance | | Empty droplet rate | Proportion of measurements that contain no detectable cellular material | Analyze blank or carrier channels for background signal | Elevated rates indicate cell dispensing problems or contamination | | Data completeness | Fraction of proteins detected across all cells in the dataset | Calculate percentage of possible protein-cell combinations observed | Low completeness complicates downstream statistical analysis | | Total ion current | Overall signal intensity of each single-cell measurement | Examine raw signal intensity distributions | Low signal suggests poor ionization or sample loss | | Mass accuracy | Precision of precursor and fragment ion mass measurements | Compare observed masses to theoretical values | Systematic errors indicate calibration problems | ## Data Inputs and Experimental Design Considerations ### Sample Preparation Methods and Their Quality Implications The method used to prepare single cells for mass spectrometry analysis has a direct impact on the quality metrics that can be achieved. The [nPOP method](https://pubmed.ncbi.nlm.nih.gov/39117766) enables parallel preparation of thousands of single cells in nanoliter-volume droplets deposited on glass slides. This approach uses a cell-dispensing and liquid-handling robot to process samples with minimal manual intervention, which reduces the variability introduced by human handling. The [nPOP protocol](https://pubmed.ncbi.nlm.nih.gov/39117766) describes two implementations with different performance characteristics. One implementation using plexDIA, a method that uses non-isobaric mass tags for data-independent acquisition, demonstrated accurate quantification of approximately 3,000 to 3,700 proteins per human cell. A separate implementation with isobaric mass tags and prioritized data acquisition enabled analysis of 1,827 single cells at a rate exceeding 1,000 single cells per day, with a depth of 800 to 1,200 proteins per human cell. These performance differences illustrate an important principle: the choice of multiplexing strategy and acquisition method affects both throughput and proteome depth. Researchers must consider whether their experimental question requires maximum protein coverage per cell or maximum cell throughput, because these goals involve tradeoffs in quality metrics. The [preprint version of the nPOP protocol](https://pubmed.ncbi.nlm.nih.gov/38076795) provides additional context on the method's flexibility and the quality control software that supports it. The QuantQC R package was developed specifically for quality control and data exploration in nPOP experiments, supporting the scaling of the method to higher plex reagents. ### Cell Number and Read Depth Considerations Experimental design decisions about cell numbers and read depth have direct consequences for data quality. The [scPerturb resource](https://pubmed.ncbi.nlm.nih.gov/38279009) provides recommendations for experimental design, including optimal cell counts and read depth, based on analysis of 44 publicly available single-cell perturbation-response datasets with molecular readouts including transcriptomics, proteomics, and epigenomics. The scPerturb work demonstrates that data interoperability is a significant challenge in single-cell perturbation studies. The researchers applied uniform quality control pipelines and harmonized feature annotations across datasets to enable development and benchmarking of computational methods. This effort highlights the importance of consistent quality control practices across experiments and laboratories. For single-cell proteomics specifically, the number of cells analyzed and the depth of sequencing per cell affect the reliability of downstream analyses. Datasets with too few cells per condition may not provide sufficient statistical power to detect meaningful differences, while datasets with insufficient read depth may miss proteins that are present at low abundance. ### Carrier and Blank Channel Design Many single-cell proteomics experiments use carrier channels containing larger amounts of protein to boost peptide identification and improve quantification accuracy. The design of these channels affects quality control metrics in important ways. Carrier channels that are too abundant can suppress signals from single-cell channels, while carrier channels that are too sparse may not provide the expected benefits for protein identification. Blank channels, which contain no cellular material, serve as negative controls for assessing background signal and contamination. The empty droplet rate, which measures the proportion of measurements containing no detectable cellular material, is an important quality metric that reflects both the performance of the cell dispensing system and the effectiveness of washing procedures between samples. ## Practical Quality Control Workflow ### Step 1: Raw Data Inspection The first stage of quality control involves examining the raw mass spectrometry data before any processing is applied. This inspection should include evaluation of total ion current across all single-cell channels, assessment of mass calibration accuracy, and identification of any obvious technical failures such as failed injections or instrument errors. The [Galaxy Training Network](https://training.galaxyproject.org/) provides accessible workflow training that can help researchers understand how to handle raw mass spectrometry data and perform initial quality assessments. These training materials emphasize reproducibility and provide practical guidance for implementing analysis workflows. During raw data inspection, researchers should record the following observations for each single-cell measurement: - Total ion current intensity and its distribution across cells - Number of features detected before filtering - Mass accuracy of precursor ions relative to theoretical values - Retention time stability across the acquisition - Any evidence of carryover between samples These observations provide the foundation for deciding whether the dataset is suitable for further processing or whether instrument or sample preparation issues need to be addressed. ### Step 2: Feature Detection and Filtering After raw data inspection, the next stage involves detecting features in the mass spectrometry data and applying initial filters. Features represent detected signals that may correspond to peptides, and filtering removes those that do not meet quality criteria such as signal intensity, mass accuracy, or chromatographic peak shape. The [nf-core documentation](https://nf-co.re/docs) describes community standards for pipeline usage and configuration that emphasize reproducibility in bioinformatics analysis. While nf-core pipelines are primarily associated with sequencing data, the principles of standardized pipeline configuration and documentation apply equally to proteomics data processing. Feature-level filtering decisions should be documented carefully, because they affect all downstream quality metrics. Common filtering criteria include: - Minimum signal intensity thresholds - Mass accuracy tolerances - Chromatographic peak width and shape criteria - Removal of features detected in blank channels The specific thresholds used should be appropriate for the instrument platform and acquisition method, and they should be applied consistently across all samples in the experiment. ### Step 3: Peptide and Protein Identification The next stage involves assigning detected features to peptide sequences and aggregating peptide measurements into protein-level quantities. This process requires searching the mass spectrometry data against a protein sequence database, which introduces additional quality considerations. The [NCBI Data Resources](https://www.ncbi.nlm.nih.gov/) provide access to sequence databases and search systems that are commonly used for protein identification in proteomics experiments. Understanding the contents and organization of these databases is important for interpreting identification results and assessing their quality. Quality metrics at this stage include: - Number of peptides identified per protein - False discovery rate estimates for peptide and protein identifications - Sequence coverage for identified proteins - Consistency of peptide identifications across replicate measurements Proteins identified by a single peptide are generally considered less reliable than those identified by multiple peptides, although single-peptide identifications may be acceptable for some applications if the peptide evidence is strong. ### Step 4: Cell-Level Quality Metrics Once peptide and protein identifications are complete, the next stage involves computing quality metrics for each individual cell in the dataset. These metrics determine which cells are retained for downstream analysis and which are excluded as low quality. The [scp workflow](https://pubmed.ncbi.nlm.nih.gov/38907155) describes quality control at the cell level as a distinct step in the data processing pipeline. Cell-level quality metrics include: - Number of proteins identified per cell - Total signal intensity per cell - Fraction of missing values across proteins - Consistency with expected cell type markers Cells with very low protein counts may represent empty droplets, failed cell lysis, or sample loss during preparation. Cells with unusually high protein counts may represent doublets or multiple cells captured in a single measurement. The [QuantQC R package](https://pubmed.ncbi.nlm.nih.gov/39117766) provides software specifically designed for quality control and data exploration in single-cell proteomics experiments. This package supports the robust scaling of nPOP to higher plex reagents and provides metrics that help researchers assess whether their data meet quality standards. ### Step 5: Normalization and Batch Correction After filtering low-quality cells, the remaining data require normalization to account for technical variation between cells and batches. Normalization adjusts protein quantities so that differences between cells reflect biological variation instead of technical artifacts. The [scp workflow](https://pubmed.ncbi.nlm.nih.gov/38907155) includes normalization and batch correction as essential steps after quality control and aggregation. These steps are particularly important in single-cell proteomics because the amount of protein in each cell can vary substantially, and this variation is not purely biological in origin. Common normalization approaches include: - Total signal normalization, which scales each cell to have the same total protein quantity - Median normalization, which aligns the median protein quantity across cells - Reference-based normalization, which uses a common reference sample to adjust for batch effects The choice of normalization method affects downstream analysis results, and researchers should evaluate the impact of different approaches on their specific dataset. ### Step 6: Final Quality Assessment Before Downstream Analysis The final stage of quality control involves a comprehensive assessment of the processed dataset before proceeding to downstream analyses such as clustering, differential expression, or pathway analysis. This assessment should integrate all the quality metrics computed at earlier stages and determine whether the dataset meets the standards required for the intended analyses. The [comparative benchmarking study of clustering algorithms](https://pubmed.ncbi.nlm.nih.gov/40903792) demonstrates that differences in data distribution, feature dimensions, and quality between different single-cell modalities pose challenges for clustering. The study evaluated 28 computational algorithms on 10 paired transcriptomic and proteomic datasets, examining performance across clustering metrics, peak memory, and running time. This benchmarking work provides actionable insights for selecting appropriate clustering approaches for specific scenarios. The findings reveal modality-specific strengths and limitations of different methods, highlighting the complementary nature of existing approaches. For top performance across both transcriptomic and proteomic data, the study recommends considering scAIDE, scDCC, and FlowSOM, with FlowSOM also offering excellent robustness. For users prioritizing memory efficiency, scDCC and scDeepCluster are recommended, while TSCAN, SHARP, and MarkovHC are recommended for users who prioritize time efficiency. The quality of the input data directly affects the performance of these clustering algorithms. Datasets with high levels of missing data, excessive technical variation, or contamination will produce unreliable clustering results regardless of the algorithm chosen. ## Records and Measurements for Quality Tracking ### Essential Records for Each Experiment Maintaining detailed records of quality control metrics is essential for tracking instrument performance over time, identifying systematic issues, and ensuring reproducibility across experiments. The following records should be maintained for each single-cell proteomics experiment: - Instrument configuration and calibration records - Sample preparation batch information - Raw data file names and acquisition parameters - Quality control metric values at each processing stage - Filtering thresholds and the number of cells retained after each filter - Software versions and parameter settings for data processing The [EMBL-EBI Training](https://www.ebi.ac.uk/training) resources provide guidance on data management practices and bioinformatics analysis education that support proper record keeping. These resources emphasize the importance of documentation for reproducible research. ### Longitudinal Quality Monitoring Quality control metrics should be tracked across experiments to identify trends that may indicate instrument degradation or systematic issues with sample preparation. For example, a gradual decline in the number of proteins identified per cell over several weeks may indicate that the mass spectrometer requires maintenance or that reagents are degrading. The [The Carpentries Lessons](https://carpentries.org/lessons) provide foundational training in data management and computing practices that support systematic quality monitoring. These lessons cover topics such as data organization, shell scripting, and version control that are valuable for implementing reproducible quality control workflows. Researchers should establish baseline quality metrics for their specific instrument and methods, then compare each new experiment against these baselines. Significant deviations from baseline values warrant investigation before proceeding with downstream analysis. ## Common Failure Patterns and Troubleshooting ### Low Protein Identification Rates One of the most common quality problems in single-cell proteomics is low protein identification rates per cell. When the number of proteins identified falls substantially below the expected range for the method being used, several potential causes should be investigated: - Sample loss during preparation, particularly during cell dispensing or transfer steps - Inefficient cell lysis or protein extraction - Insufficient digestion of proteins into peptides - Contamination that suppresses ionization efficiency - Instrument sensitivity issues or calibration drift The [nPOP protocol](https://pubmed.ncbi.nlm.nih.gov/39117766) provides detailed guidance on sample preparation steps that affect protein identification rates. The method's emphasis on nanoliter-volume droplets and minimal sample handling is designed to reduce losses that commonly occur with conventional sample preparation approaches. ### High Coefficient of Variation Excessive quantitative variation between technical replicates or similar cells indicates problems with reproducibility. Potential causes include: - Inconsistent sample preparation between cells - Variation in digestion efficiency - Ionization suppression from contaminants - Instrument instability during acquisition - Incomplete mixing of reagents in nanoliter volumes When coefficient of variation values are high, researchers should examine whether the variation is concentrated in specific proteins or distributed across the entire proteome. Protein-specific variation may indicate biological heterogeneity, while global variation suggests technical issues. ### Elevated Empty Droplet Rates Empty droplets are measurements that contain no detectable cellular material. While some empty droplets are expected in any single-cell experiment, elevated rates indicate problems with cell dispensing or sample preparation. The [nPOP method](https://pubmed.ncbi.nlm.nih.gov/38076795) uses a cell-dispensing robot to deposit single cells into nanoliter-volume droplets. The accuracy of this dispensing process directly affects empty droplet rates. Factors that can increase empty droplet rates include: - Low cell viability in the suspension - Cell settling or clumping in the dispensing system - Clogging of the dispensing nozzle - Inaccurate cell counting before dispensing When empty droplet rates are elevated, researchers should examine the cell suspension quality and the dispensing system performance before proceeding with data analysis. ### Batch Effects and Systematic Variation Single-cell proteomics experiments often involve multiple batches of samples processed at different times. Systematic differences between batches can confound biological comparisons and must be addressed through appropriate normalization and batch correction. The [scPerturb resource](https://pubmed.ncbi.nlm.nih.gov/38279009) addresses the challenge of data interoperability across single-cell perturbation datasets by applying uniform quality control pipelines and harmonizing feature annotations. This work demonstrates that batch effects are a significant challenge in single-cell studies and that consistent quality control practices are essential for comparing data across experiments. Researchers should design experiments to minimize batch effects where possible, such as by randomizing sample processing order and including bridge samples that are analyzed in multiple batches. When batch effects are unavoidable, statistical methods for batch correction should be applied after quality control and normalization. ## Limitations of Quality Control Metrics ### Threshold Values Are Method-Specific The quality control thresholds described in this article are general guidelines instead of universal standards. The expected number of proteins per cell, acceptable coefficient of variation, and tolerable empty droplet rates depend on the specific method used, the instrument platform, and the biological sample being analyzed. The [nPOP protocol](https://pubmed.ncbi.nlm.nih.gov/39117766) demonstrates this principle by describing two implementations with different performance characteristics. The plexDIA implementation achieved approximately 3,000 to 3,700 proteins per human cell, while the isobaric tag implementation achieved 800 to 1,200 proteins per cell. Both implementations are considered successful, but they have different quality profiles that reflect their different design goals. Researchers should establish method-specific quality thresholds based on their own experience and published benchmarks for their specific approach. Comparing quality metrics across different methods without accounting for these differences can lead to incorrect conclusions about data quality. ### Missing Data and Its Consequences Single-cell proteomics datasets typically contain substantial amounts of missing data, because many proteins are not detected in every cell. This missingness is not random but is influenced by protein abundance, detection limits, and technical factors. The [comparative benchmarking study](https://pubmed.ncbi.nlm.nih.gov/40903792) discusses the impact of highly variable genes and cell type granularity on clustering performance, and the study used simulated datasets to evaluate the robustness of clustering methods. These considerations are relevant to quality control because the amount and pattern of missing data affect the reliability of downstream analyses. Researchers should assess the pattern of missing data in their datasets and consider whether imputation methods are appropriate. The choice of imputation approach can substantially affect downstream results, and different approaches are appropriate for different missing data mechanisms. ### Quality Metrics Cannot Detect All Problems Quality control metrics provide valuable information about data quality, but they cannot detect all potential problems. Some issues may produce datasets that pass all standard quality checks but still contain systematic errors that affect biological conclusions. For example, contamination with proteins from other species or from the sample preparation environment may not be detected by standard quality metrics if the contaminating proteins are consistently present across all cells. Similarly, subtle changes in protein quantification that result from incomplete digestion or modification during sample preparation may not be apparent from quality metrics alone. The [scp workflow](https://pubmed.ncbi.nlm.nih.gov/38907155) addresses this limitation by using ground truth datasets with known composition for benchmarking. The developers created a design with cell lines mixed in known proportions to generate controlled variability, allowing them to validate their workflow and identify potential issues that might not be apparent from quality metrics alone. ## Professional Escalation Criteria ### When to Seek Expert Assistance Certain quality control findings indicate problems that require expert assistance to resolve. Researchers should escalate to instrument manufacturers, bioinformatics specialists, or experienced colleagues when they encounter the following situations: - Persistent low protein identification rates that do not improve after troubleshooting sample preparation - Sudden changes in quality metrics that suggest instrument malfunction - Quality control failures that cannot be explained by sample preparation or acquisition parameters - The need to establish quality control procedures for a new method or instrument platform The [EMBL-EBI Training](https://www.ebi.ac.uk/training) resources provide pathways for developing bioinformatics skills that may help researchers address quality control challenges independently. However, some problems require specialized expertise that is best obtained through collaboration with experienced practitioners. ### Documentation for Escalation When escalating quality control problems, researchers should provide comprehensive documentation that includes: - Detailed description of the quality control metrics that failed - Raw data files and processing parameters - Sample preparation records and reagent information - Instrument maintenance and calibration history - Any troubleshooting steps already attempted This documentation enables experts to diagnose problems more efficiently and reduces the time required to resolve quality issues. ## Safety and Regulatory Context ### Data Integrity and Reproducibility Quality control in single-cell proteomics is closely connected to data integrity and reproducibility. The [nf-core documentation](https://nf-co.re/docs) emphasizes community standards for pipeline usage and configuration that support reproducible analysis. These standards include version control, containerization, and comprehensive documentation of analysis parameters. Researchers should ensure that their quality control procedures are documented sufficiently to allow other researchers to reproduce their analyses. This documentation should include specific thresholds used, software versions, and any manual decisions made during the quality control process. ### Ethical Considerations in Data Reporting Quality control decisions affect which data are reported and interpreted. Researchers have an ethical obligation to report quality control procedures transparently and to avoid selective reporting that could bias conclusions. The [scPerturb resource](https://pubmed.ncbi.nlm.nih.gov/38279009) demonstrates the value of harmonized quality control pipelines for enabling comparison across datasets. By applying uniform quality control procedures, the researchers made it possible for other investigators to benchmark computational methods and compare results across studies. Researchers should report the number of cells excluded during quality control, the reasons for exclusion, and the thresholds used. This transparency allows readers to assess the potential impact of quality filtering on the reported results. ## Decision Framework for Go or No-Go Analysis in Single-Cell Proteomics ### The Need for a Structured Decision Process Quality control metrics provide measurements, but researchers still face the challenge of translating those measurements into a clear decision about whether to proceed with downstream analysis. A structured decision framework transforms raw quality metrics into actionable choices, reducing the risk of proceeding with datasets that will produce misleading biological conclusions or discarding datasets that could yield valuable information with appropriate adjustments. The [standardized workflow using the scp package](https://pubmed.ncbi.nlm.nih.gov/38907155) emphasizes that data analysis approaches differ substantially between research teams and that evaluating pipelines is difficult because ground truth datasets are often missing. This variability in practice creates a need for a transparent, reproducible decision process that can be applied consistently across experiments and laboratories. A go or no-go decision framework addresses this need by defining explicit criteria for proceeding, adjusting, or halting analysis. The framework should be established before data collection begins, documented in the analysis plan, and applied consistently to all datasets generated under the same experimental conditions. ### Tiered Decision Categories The decision framework uses three categories instead of a simple binary pass or fail outcome. This tiered approach recognizes that quality problems exist on a continuum and that some issues can be addressed through computational correction while others require repeating the experiment. **Tier 1: Proceed with Analysis** Datasets in this category meet all primary quality thresholds and are suitable for downstream analysis without major adjustments. The criteria for this tier include protein identification rates within the expected range for the method, acceptable quantitative reproducibility, empty droplet rates below the established threshold, and data completeness sufficient for the planned statistical analyses. For datasets in this tier, researchers should proceed with normalization, batch correction, and downstream analysis while continuing to monitor quality metrics throughout the processing pipeline. The [scp workflow](https://pubmed.ncbi.nlm.nih.gov/38907155) describes quality control at both the feature level and the cell level, followed by aggregation of raw data into peptides and proteins, normalization, and batch correction. This multistage approach ensures that quality is maintained throughout the processing pipeline. **Tier 2: Proceed with Adjustments** Datasets in this category have quality metrics that fall outside the ideal range but can be addressed through computational adjustments or selective filtering. Examples include datasets with moderate levels of missing data that can be handled through appropriate imputation methods, datasets with batch effects that can be corrected statistically, or datasets where a subset of low-quality cells can be excluded without compromising the overall experimental design. The [comparative benchmarking study of clustering algorithms](https://pubmed.ncbi.nlm.nih.gov/40903792) provides guidance relevant to this tier. The study evaluated 28 computational algorithms on 10 paired transcriptomic and proteomic datasets and found that different methods have modality-specific strengths and limitations. For datasets with quality issues, the choice of downstream algorithm becomes more important, and researchers should select methods that are robust to the specific quality problems present in their data. When proceeding with adjustments, researchers must document the adjustments made, the rationale for those adjustments, and the potential impact on downstream results. This documentation is essential for interpreting results and for enabling other researchers to evaluate the validity of the conclusions. **Tier 3: Do Not Proceed** Datasets in this category have quality problems that cannot be addressed through computational adjustments alone. Examples include widespread sample loss during preparation, instrument malfunction during acquisition, contamination that affects a substantial portion of measurements, or empty droplet rates so high that the remaining cells do not provide sufficient statistical power. The [nPOP protocol](https://pubmed.ncbi.nlm.nih.gov/39117766) describes methods for preparing thousands of single cells in nanoliter-volume droplets, with implementations achieving analysis of over 1,000 single cells per day. When a substantial fraction of these measurements fail quality checks, the cost of proceeding with analysis may outweigh the value of the information obtained. For datasets in this tier, researchers should halt analysis, investigate the root cause of the quality problems, and determine whether the experiment should be repeated with adjusted protocols. The decision to repeat an experiment should consider the cost of additional sample preparation and instrument time against the value of obtaining reliable data. ### Establishing Thresholds for Each Tier The thresholds that define each tier must be established before data collection begins and should be based on the specific method, instrument platform, and biological sample being analyzed. The [nPOP protocol](https://pubmed.ncbi.nlm.nih.gov/39117766) demonstrates that different implementations achieve different performance characteristics, with the plexDIA implementation achieving approximately 3,000 to 3,700 proteins per human cell and the isobaric tag implementation achieving 800 to 1,200 proteins per cell. Researchers should establish thresholds based on published benchmarks for their specific method, their own experience with the method, and the requirements of their planned downstream analyses. The thresholds should be documented in the analysis plan and applied consistently across all datasets. For protein identification rates, the threshold for Tier 1 should be set at or near the expected range for the method. The threshold for Tier 2 should be set at a level where the reduced protein coverage still allows the planned analyses to be performed with acceptable reliability. The threshold for Tier 3 should be set at a level where the protein coverage is insufficient for meaningful biological interpretation. For empty droplet rates, the threshold for Tier 1 should reflect the expected performance of the cell dispensing system. The [nPOP method](https://pubmed.ncbi.nlm.nih.gov/38076795) uses a cell-dispensing robot to deposit single cells into nanoliter-volume droplets, and the accuracy of this dispensing process directly affects empty droplet rates. The threshold for Tier 2 should allow for some elevation in empty droplet rates while still retaining sufficient cells for analysis. The threshold for Tier 3 should reflect a level where the remaining cells are too few for reliable statistical analysis. For coefficient of variation, the threshold for Tier 1 should reflect the expected technical variability for the method. The threshold for Tier 2 should allow for moderate increases in variability that can be addressed through normalization or by focusing on proteins with more consistent measurements. The threshold for Tier 3 should reflect a level where quantitative comparisons between cells are unlikely to be reliable. ### Implementing the Decision Framework The decision framework should be implemented as a formal process with documented criteria and recorded outcomes. The following steps describe how to implement the framework in practice. **Step 1: Define thresholds before data collection** Establish the specific thresholds for each quality metric and each tier before beginning the experiment. Document these thresholds in the analysis plan along with the rationale for each threshold. The [EMBL-EBI Training](https://www.ebi.ac.uk/training) resources provide guidance on data management practices that support this type of advance planning. **Step 2: Compute quality metrics after data processing** After completing the initial data processing steps, compute all quality metrics defined in the quality control workflow. The [QuantQC R package](https://pubmed.ncbi.nlm.nih.gov/39117766) provides software specifically designed for quality control and data exploration in single-cell proteomics experiments, supporting the robust scaling of nPOP to higher plex reagents. **Step 3: Apply the tier criteria** Compare the computed quality metrics against the predefined thresholds and assign the dataset to a tier. This assignment should be done systematically, with each quality metric evaluated against its threshold and the overall tier determined by the most restrictive metric. **Step 4: Document the decision** Record the tier assignment, the quality metric values that led to the assignment, and any adjustments made in response to the assignment. This documentation should be maintained with the experimental records and referenced in any publications resulting from the analysis. **Step 5: Review and refine thresholds** After completing several experiments, review the threshold values and refine them based on accumulated experience. The [scPerturb resource](https://pubmed.ncbi.nlm.nih.gov/38279009) demonstrates the value of harmonized quality control pipelines for enabling comparison across datasets, and similar principles apply to refining thresholds within a single laboratory. ### Common Decision Scenarios Several common scenarios illustrate how the decision framework operates in practice. **Scenario 1: Marginal protein identification rates** A dataset shows protein identification rates slightly below the Tier 1 threshold but above the Tier 2 threshold. The decision framework assigns this dataset to Tier 2, allowing analysis to proceed with adjustments. The researcher should evaluate whether the reduced protein coverage affects the specific biological questions being addressed and consider whether additional filtering or imputation is needed. **Scenario 2: Elevated empty droplet rates in one batch** A multi-batch experiment shows elevated empty droplet rates in one batch but acceptable rates in other batches. The decision framework assigns the affected batch to Tier 2 or Tier 3 depending on the severity of the elevation. If the affected batch can be excluded without compromising the experimental design, the researcher may proceed with analysis of the remaining batches. If the affected batch is essential to the experimental design, the experiment may need to be repeated. **Scenario 3: High coefficient of variation with no identifiable cause** A dataset shows high coefficient of variation that cannot be explained by sample preparation or instrument issues. The decision framework assigns this dataset to Tier 2 or Tier 3 depending on the severity of the variation. The [scp workflow](https://pubmed.ncbi.nlm.nih.gov/38907155) addresses this challenge by using ground truth datasets with known composition for benchmarking, allowing researchers to evaluate whether their processing pipeline introduces variability that could be reduced through different processing choices. ### Integrating the Decision Framework with Existing Workflows The decision framework should be integrated with existing quality control workflows instead of replacing them. The [scp workflow](https://pubmed.ncbi.nlm.nih.gov/38907155) describes quality control at the feature level and the cell level, followed by aggregation of raw data into peptides and proteins, normalization, and batch correction. The decision framework adds a formal decision point after these quality control steps, providing a structured way to determine whether the processed data are suitable for downstream analysis. The [nf-core documentation](https://nf-co.re/docs) describes community standards for pipeline usage and configuration that emphasize reproducibility in bioinformatics analysis. The decision framework aligns with these principles by providing a documented, reproducible process for quality assessment and decision making. The [Galaxy Training Network](https://training.galaxyproject.org/) provides accessible workflow training that can help researchers implement the decision framework as part of their analysis pipelines. These training materials emphasize reproducibility and provide practical guidance for implementing analysis workflows. ### Limitations of the Decision Framework The decision framework provides structure but cannot eliminate all uncertainty in quality assessment. Several limitations should be acknowledged. The thresholds used in the framework are method-specific and must be established based on experience and published benchmarks. The [nPOP protocol](https://pubmed.ncbi.nlm.nih.gov/39117766) demonstrates that different implementations achieve different performance characteristics, and thresholds that are appropriate for one method may not be appropriate for another. The framework cannot detect all quality problems. Some issues may produce datasets that pass all quality checks but still contain systematic errors that affect biological conclusions. The [scp workflow](https://pubmed.ncbi.nlm.nih.gov/38907155) addresses this limitation by using ground truth datasets with known composition for benchmarking, allowing researchers to validate their processing pipeline and identify potential issues that might not be apparent from quality metrics alone. The framework requires judgment in its application. While the thresholds are defined in advance, the interpretation of borderline cases and the decision to proceed with adjustments require scientific judgment. This judgment should be documented and reviewed as part of the quality control process. ### Professional Escalation Within the Decision Framework The decision framework should include explicit criteria for professional escalation. When a dataset falls into Tier 3, or when a dataset falls into Tier 2 but the required adjustments are complex or uncertain, researchers should seek expert assistance. The [EMBL-EBI Training](https://www.ebi.ac.uk/training) resources provide pathways for developing bioinformatics skills that may help researchers address quality control challenges independently. However, some problems require specialized expertise that is best obtained through collaboration with experienced practitioners. When escalating quality control problems, researchers should provide comprehensive documentation that includes the quality control metrics that failed, the raw data files and processing parameters, sample preparation records and reagent information, instrument maintenance and calibration history, and any troubleshooting steps already attempted. This documentation enables experts to diagnose problems more efficiently and reduces the time required to resolve quality issues. The [The Carpentries Lessons](https://carpentries.org/lessons) provide foundational training in data management and computing practices that support systematic quality monitoring and documentation. These lessons cover topics such as data organization, shell scripting, and version control that are valuable for implementing reproducible quality control workflows and for preparing documentation for professional escalation. ## Frequently Asked Questions ### What is the minimum number of proteins per cell for acceptable single-cell proteomics data? The minimum acceptable number of proteins per cell depends on the method used and the biological question being addressed. The [nPOP protocol](https://pubmed.ncbi.nlm.nih.gov/39117766) describes implementations achieving approximately 3,000 to 3,700 proteins per human cell with plexDIA and 800 to 1,200 proteins per cell with isobaric tags. Researchers should establish thresholds based on their specific method and compare their results to published benchmarks for that approach. ### How is the empty droplet rate calculated and what is an acceptable level? The empty droplet rate is calculated as the proportion of single-cell measurements that contain no detectable cellular material, typically identified by very low protein counts or signal intensities. Acceptable levels depend on the cell dispensing method and the biological sample. The [nPOP method](https://pubmed.ncbi.nlm.nih.gov/38076795) uses robotic cell dispensing to minimize empty droplets, but some empty measurements are expected in any single-cell experiment. ### What coefficient of variation should I expect for technical replicates in single-cell proteomics? Coefficient of variation values depend on the specific method, instrument platform, and protein abundance range. The [scp workflow](https://pubmed.ncbi.nlm.nih.gov/38907155) uses controlled variability datasets with cell lines mixed in known proportions to benchmark data processing approaches. Researchers should evaluate coefficient of variation for their specific method and compare against published values for similar approaches. ### How should I handle cells with very low protein counts in my dataset? Cells with very low protein counts may represent empty droplets, failed cell lysis, or sample loss. These cells should be examined carefully before exclusion. The [QuantQC R package](https://pubmed.ncbi.nlm.nih.gov/39117766) provides metrics and software for quality control that support decisions about which cells to retain. Researchers should document the criteria used for cell exclusion and report the number of cells removed. ### Can I compare quality metrics across different single-cell proteomics methods? Direct comparison of quality metrics across different methods is challenging because methods have different design goals and performance characteristics. The [nPOP protocol](https://pubmed.ncbi.nlm.nih.gov/39117766) demonstrates that different implementations achieve different balances of throughput and proteome depth. Researchers should compare their results to benchmarks for their specific method instead of to results from different approaches. ### What should I do if my quality control metrics are consistently below published benchmarks? Consistently poor quality metrics warrant systematic investigation of sample preparation, instrument performance, and data processing parameters. The [EMBL-EBI Training](https://www.ebi.ac.uk/training) resources provide education on bioinformatics analysis that may help identify processing issues. If problems persist after troubleshooting, consultation with instrument manufacturers or experienced practitioners is recommended. ### How does data completeness affect downstream clustering analysis? Data completeness, or the fraction of proteins detected across all cells, directly affects clustering performance. The [comparative benchmarking study](https://pubmed.ncbi.nlm.nih.gov/40903792) evaluated 28 clustering algorithms on paired transcriptomic and proteomic datasets and found that data quality differences between modalities pose challenges for clustering. Datasets with substantial missing data may require imputation or specialized algorithms designed to handle missing values. ### What quality control metrics should I report in my publications? Publications should report the number of cells analyzed, the number excluded during quality control and the reasons for exclusion, the median or mean number of proteins identified per cell, and the thresholds used for quality filtering. The [scPerturb resource](https://pubmed.ncbi.nlm.nih.gov/38279009) demonstrates the value of uniform quality control pipelines for enabling data comparison across studies. Transparent reporting of quality control procedures supports reproducibility and allows readers to assess data quality. ## Related Bioinformatics Guides - [Proteomics Mass Spectrometry: From Sample Preparation to Data Analysis](/knowledge/bioinformatics/proteomics-mass-spectrometry-from-sample-preparation-to-data-analysis) - [Mass Spectrometry-Based Proteomics: Data Analysis Pipelines and Tools](/knowledge/bioinformatics/mass-spectrometry-based-proteomics-data-analysis-pipelines-and-tools) - [Single-Cell RNA Sequencing Quality Control: A Practical Guide to Filtering and Metrics](/knowledge/bioinformatics/single-cell-rna-sequencing-quality-control-a-practical-guide-to-filtering-and-metrics) - [Single-Cell Sequencing Workflow: From Sample Preparation to Data Analysis](/knowledge/bioinformatics/single-cell-sequencing-workflow-from-sample-preparation-to-data-analysis) - [Single-Cell Sequencing Analysis Pipeline: From Raw Data to Biological Insights](/knowledge/bioinformatics/single-cell-sequencing-analysis-pipeline-from-raw-data-to-biological-insights) ## References and Further Reading - [NCBI Data Resources](https://www.ncbi.nlm.nih.gov/). National Center for Biotechnology Information. - [EMBL-EBI Training](https://www.ebi.ac.uk/training). European Bioinformatics Institute. - [Bioconductor](https://bioconductor.org/). Bioconductor Project. - [Galaxy Training Network](https://training.galaxyproject.org/). Galaxy Project. - [nf-core Documentation](https://nf-co.re/docs). nf-core. - [The Carpentries Lessons](https://carpentries.org/lessons). The Carpentries. - [scPerturb: harmonized single-cell perturbation data.](https://pubmed.ncbi.nlm.nih.gov/38279009). Nature methods, 2024. - [Massively parallel sample preparation for multiplexed single-cell proteomics using nPOP.](https://pubmed.ncbi.nlm.nih.gov/39117766). Nature protocols, 2024. - [Massively parallel sample preparation for multiplexed single-cell proteomics using nPOP.](https://pubmed.ncbi.nlm.nih.gov/38076795). bioRxiv : the preprint server for biology, 2023. - [Standardized Workflow for Mass-Spectrometry-Based Single-Cell Proteomics Data Processing and Analysis Using the scp Package.](https://pubmed.ncbi.nlm.nih.gov/38907155). Methods in molecular biology (Clifton, N.J.), 2024. - [Comparative benchmarking of single-cell clustering algorithms for transcriptomic and proteomic data.](https://pubmed.ncbi.nlm.nih.gov/40903792). Genome biology, 2025. > This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.