Rather than relying on universal cutoffs, researchers should use QC metrics to distinguish genuine biological variation from technical artifacts.<\/p>\n\n\n\n
Understanding the Sources of Technical Noise<\/strong><\/h2>\n\n\n\nBefore selecting thresholds, it is important to understand what the QC metrics actually measure.<\/p>\n\n\n\n
1. Ambient RNA and Lateral Leakage<\/strong><\/h3>\n\n\n\nDuring tissue processing, transcripts released from damaged cells may diffuse across the tissue section before capture. These ambient transcripts create low-level background expression in regions containing few or no intact cells.<\/p>\n\n\n\n
Because ambient RNA is typically distributed broadly rather than confined to cellular boundaries, affected cells often exhibit low transcript density and weak expression across many genes instead of strong expression of biologically coherent markers.<\/p>\n\n\n\n![\"\"](\"https:\/\/bioturing.com\/blog\/wp-content\/uploads\/2026\/09\/image-1-1024x575.jpeg\")
<\/figure>\n\n\n\nFigure 1<\/strong>: Conceptual illustration of ambient RNA and mixed cellular contributions in sequencing-based spatial transcriptomics. RNA released from damaged or disrupted cells can contribute background transcripts to neighboring capture regions, while individual spatial capture units may also contain transcripts originating from multiple cells. This conceptual figure was generated using Canvas AI.<\/p>\n\n\n\n2. Segmentation Artifacts<\/strong><\/h3>\n\n\n\nCell segmentation is never perfect.<\/p>\n\n\n\n
Small segmented objects may represent cell fragments, debris, or incomplete nuclei, whereas unusually large objects may result from the merging of neighboring cells (doublets). In both cases, segmentation errors can distort downstream analyses because the measured molecular profile may no longer represent a single biological cell.<\/p>\n\n\n\n
Examining cell area together with transcript density<\/strong> can provide complementary information for assessing segmentation quality and is an important step before downstream analysis 1<\/sup>. For example, Mitchel et al. (2026) 2<\/sup> systematically evaluated the impact of segmentation errors across multiple spatial transcriptomics platforms<\/a>, including Xenium, CosMx, and MERFISH, and across different tissue types and organisms. Their study demonstrated that segmentation errors can lead to molecular misassignment between neighboring cells and substantially affect downstream analyses, including differential expression (DE), cell\u2013cell interaction analyses, and inference of cellular states. These findings highlight the importance of carefully evaluating computational results in the context of biological knowledge and experimental design, rather than relying solely on computational outputs.<\/p>\n\n\n\n3. Mitochondrial Transcripts<\/strong><\/h3>\n\n\n\nMitochondrial RNA is commonly used as an indicator of cellular stress 3<\/sup>.<\/p>\n\n\n\nCells undergoing apoptosis or membrane damage lose cytoplasmic RNA while retaining mitochondrial transcripts, increasing the fraction of mitochondrial gene expression. However, mitochondrial content varies naturally between tissues and cell types. Cardiomyocytes, neurons, and metabolically active tumor cells may exhibit elevated mitochondrial expression despite being biologically healthy.<\/p>\n\n\n\n
For this reason, mitochondrial percentage should be interpreted in the context of the overall dataset rather than filtered using a fixed threshold.<\/p>\n\n\n\n
4. Negative Control Probes<\/strong><\/h3>\n\n\n\nModern imaging assays include negative-control probes that should not hybridize to endogenous transcripts 4<\/sup>.<\/p>\n\n\n\nThese probes provide an estimate of optical background and nonspecific signal. Cells with unusually high negative-control counts are more likely to represent technical artifacts than genuine biological observations.<\/p>\n\n\n\n
Within SpatialX, transcripts matching keywords such as NEGPRB<\/strong>, BLANK<\/strong>, CONTROL<\/strong>, or UNASSIGNED<\/strong> are automatically recognized, allowing researchers to visualize background signals alongside biological measurements.<\/p>\n\n\n\nBuilding Data-Driven QC Thresholds<\/strong><\/h2>\n\n\n\nThe goal of quality control (QC) is not simply to remove low-quality cells; rather, it is to preserve biologically meaningful variation while excluding observations that cannot be interpreted with confidence.<\/p>\n\n\n\n
Instead of applying arbitrary cutoffs, researchers should first examine the distribution of each QC metric across the dataset. This allows thresholds to be defined based on the characteristics of the experiment rather than relying on predefined values.<\/p>\n\n\n\n
To evaluate QC metrics effectively, computational biologists can visualize transcript counts, the number of detected genes, mitochondrial percentage, negative-control signal, cell area, and transcript density before applying filters. Such visualizations can reveal outliers and systematic differences between samples, providing a data-driven basis for selecting appropriate thresholds. For spatial datasets, these metrics can be further evaluated using tools such as SpatialQC 5<\/sup>, which provides a framework for assessing key QC features (Figure 2).<\/p>\n\n\n\n![\"\"](\"https:\/\/bioturing.com\/blog\/wp-content\/uploads\/2026\/09\/image.jpeg\")
<\/figure>\n\n\n\nFigure 2<\/strong>: Common QC workflow illustrated in SpatialQC by Mao et al. (2024) 5<\/sup>. Key QC metrics, including mitochondrial percentage and gene-level counts, can be used to assess data quality. When data from multiple slides are integrated, it is also important to compare QC features across datasets and identify potential slide-specific differences or batch effects. The figure is adapted from Mao et al. (2024) 5<\/sup> and is distributed under the Creative Commons Attribution License.<\/p>\n\n\n\nInteractive Threshold Optimization in SpatialX<\/strong><\/h2>\n\n\n\nMany QC decisions require iterative exploration.<\/p>\n\n\n\n
SpatialX allows researchers to adjust thresholds interactively while immediately observing their impact on both tissue morphology and downstream analyses.<\/p>\n\n\n\n
Rather than asking “Should I filter cells with more than 10% mitochondrial RNA?”<\/em>, researchers can instead ask:<\/p>\n\n\n\n\n- Which cells are removed at this threshold \u2014 and does the answer change if I filter on % negative control or transcripts\/\u03bcm\u00b2 instead of mito%?<\/li>\n\n\n\n
- Where are these cells located in the tissue?<\/li>\n\n\n\n
- Do they cluster in damaged regions, or do they mark a real population (e.g., high-mito cells in a metabolically active zone)?<\/li>\n\n\n\n
- After sub-clustering the retained cells, does filtering sharpen cluster separation or just remove data?<\/li>\n<\/ul>\n\n\n\n
SpatialX supports this at two points in the workflow: threshold-based filtering at submission (per sample or applied uniformly across a multi-sample study), and post-hoc QC on existing studies, where thresholds are applied to stored QC metadata and filtered cells can be sub-clustered immediately to test the effect \u2014 before any data is permanently removed.<\/p>\n\n\n\n
Interpreting Common QC Metrics<\/strong><\/h2>\n\n\n\nEvery QC decision comes down to reading four signals correctly: cell area, transcript density, mitochondrial percentage, and negative-control signal. Each one flags a different failure mode \u2014 segmentation errors, ambient contamination, cellular stress, background noise \u2014 and learning to read them individually is what makes joint interpretation possible later. <\/p>\n\n\n\n
1. Cell Area<\/strong><\/h3>\n\n\n\nVery small segmented regions frequently represent debris or incomplete segmentation.<\/p>\n\n\n\n
Very large regions often indicate merged neighboring cells (doublets).<\/p>\n\n\n\n
Examining the distribution of cell areas helps identify segmentation errors while preserving naturally large cell types.<\/p>\n\n\n\n
2. Transcript Density<\/strong><\/h3>\n\n\n\nTranscript density normalizes transcript counts by segmented cell area.<\/p>\n\n\n\n
Unlike total transcript count, density distinguishes genuinely transcriptionally active cells from large, poorly segmented regions containing mostly background signals. Low-density objects frequently correspond to ambient RNA contamination or segmentation artifacts, making this one of the most informative QC metrics for imaging-based spatial assays.<\/p>\n\n\n\n
3. Mitochondrial Percentage<\/strong><\/h3>\n\n\n\nRather than applying a universal cutoff, mitochondrial expression should be evaluated relative to the tissue under study.<\/p>\n\n\n\n
Visualizing mitochondrial percentage together with clustering often reveals whether high-mitochondrial cells represent stressed populations or biologically distinct cell types that should be retained.<\/p>\n\n\n\n
4. Negative-Control Signal<\/strong><\/h3>\n\n\n\nNegative-control probes estimate the experimental noise floor.<\/p>\n\n\n\n
Cells with elevated negative-control signal should be examined alongside transcript density and total transcript counts before deciding whether they represent technical artifacts.<\/p>\n\n\n\n
Together, these four metrics form a diagnostic framework for spatial data: area identifies segmentation errors, density distinguishes true transcriptional signal from background, mitochondrial content indicates whether cells are metabolically compromised or represent a distinct biological population, and negative-control probes quantify the baseline rate of non-specific signal. <\/p>\n\n\n\n
Interpreting them jointly \u2014 rather than any one alone \u2014 is what allows QC to reflect the actual condition of the tissue, rather than an arbitrary cutoff. <\/p>\n\n\n\n
From Quality Control to Biological Discovery<\/strong><\/h2>\n\n\n\nHigh-quality QC improves every downstream analysis. Removing technical artifacts leads to more stable clustering, clearer cell-type boundaries, more reliable differential expression, and more accurate spatial neighborhood and ligand-receptor analyses. Importantly, thoughtful QC reduces false biological discoveries while preserving rare but meaningful cellular populations.<\/p>\n\n\n\n
During tissue processing, transcripts released from damaged cells may diffuse across the tissue section before capture. These ambient transcripts create low-level background expression in regions containing few or no intact cells.<\/p>\n\n\n\n
Because ambient RNA is typically distributed broadly rather than confined to cellular boundaries, affected cells often exhibit low transcript density and weak expression across many genes instead of strong expression of biologically coherent markers.<\/p>\n\n\n\n Figure 1<\/strong>: Conceptual illustration of ambient RNA and mixed cellular contributions in sequencing-based spatial transcriptomics. RNA released from damaged or disrupted cells can contribute background transcripts to neighboring capture regions, while individual spatial capture units may also contain transcripts originating from multiple cells. This conceptual figure was generated using Canvas AI.<\/p>\n\n\n\n Cell segmentation is never perfect.<\/p>\n\n\n\n Small segmented objects may represent cell fragments, debris, or incomplete nuclei, whereas unusually large objects may result from the merging of neighboring cells (doublets). In both cases, segmentation errors can distort downstream analyses because the measured molecular profile may no longer represent a single biological cell.<\/p>\n\n\n\n Examining cell area together with transcript density<\/strong> can provide complementary information for assessing segmentation quality and is an important step before downstream analysis 1<\/sup>. For example, Mitchel et al. (2026) 2<\/sup> systematically evaluated the impact of segmentation errors across multiple spatial transcriptomics platforms<\/a>, including Xenium, CosMx, and MERFISH, and across different tissue types and organisms. Their study demonstrated that segmentation errors can lead to molecular misassignment between neighboring cells and substantially affect downstream analyses, including differential expression (DE), cell\u2013cell interaction analyses, and inference of cellular states. These findings highlight the importance of carefully evaluating computational results in the context of biological knowledge and experimental design, rather than relying solely on computational outputs.<\/p>\n\n\n\n Mitochondrial RNA is commonly used as an indicator of cellular stress 3<\/sup>.<\/p>\n\n\n\n Cells undergoing apoptosis or membrane damage lose cytoplasmic RNA while retaining mitochondrial transcripts, increasing the fraction of mitochondrial gene expression. However, mitochondrial content varies naturally between tissues and cell types. Cardiomyocytes, neurons, and metabolically active tumor cells may exhibit elevated mitochondrial expression despite being biologically healthy.<\/p>\n\n\n\n For this reason, mitochondrial percentage should be interpreted in the context of the overall dataset rather than filtered using a fixed threshold.<\/p>\n\n\n\n Modern imaging assays include negative-control probes that should not hybridize to endogenous transcripts 4<\/sup>.<\/p>\n\n\n\n These probes provide an estimate of optical background and nonspecific signal. Cells with unusually high negative-control counts are more likely to represent technical artifacts than genuine biological observations.<\/p>\n\n\n\n Within SpatialX, transcripts matching keywords such as NEGPRB<\/strong>, BLANK<\/strong>, CONTROL<\/strong>, or UNASSIGNED<\/strong> are automatically recognized, allowing researchers to visualize background signals alongside biological measurements.<\/p>\n\n\n\n The goal of quality control (QC) is not simply to remove low-quality cells; rather, it is to preserve biologically meaningful variation while excluding observations that cannot be interpreted with confidence.<\/p>\n\n\n\n Instead of applying arbitrary cutoffs, researchers should first examine the distribution of each QC metric across the dataset. This allows thresholds to be defined based on the characteristics of the experiment rather than relying on predefined values.<\/p>\n\n\n\n To evaluate QC metrics effectively, computational biologists can visualize transcript counts, the number of detected genes, mitochondrial percentage, negative-control signal, cell area, and transcript density before applying filters. Such visualizations can reveal outliers and systematic differences between samples, providing a data-driven basis for selecting appropriate thresholds. For spatial datasets, these metrics can be further evaluated using tools such as SpatialQC 5<\/sup>, which provides a framework for assessing key QC features (Figure 2).<\/p>\n\n\n\n Figure 2<\/strong>: Common QC workflow illustrated in SpatialQC by Mao et al. (2024) 5<\/sup>. Key QC metrics, including mitochondrial percentage and gene-level counts, can be used to assess data quality. When data from multiple slides are integrated, it is also important to compare QC features across datasets and identify potential slide-specific differences or batch effects. The figure is adapted from Mao et al. (2024) 5<\/sup> and is distributed under the Creative Commons Attribution License.<\/p>\n\n\n\n Many QC decisions require iterative exploration.<\/p>\n\n\n\n SpatialX allows researchers to adjust thresholds interactively while immediately observing their impact on both tissue morphology and downstream analyses.<\/p>\n\n\n\n Rather than asking “Should I filter cells with more than 10% mitochondrial RNA?”<\/em>, researchers can instead ask:<\/p>\n\n\n\n SpatialX supports this at two points in the workflow: threshold-based filtering at submission (per sample or applied uniformly across a multi-sample study), and post-hoc QC on existing studies, where thresholds are applied to stored QC metadata and filtered cells can be sub-clustered immediately to test the effect \u2014 before any data is permanently removed.<\/p>\n\n\n\n Every QC decision comes down to reading four signals correctly: cell area, transcript density, mitochondrial percentage, and negative-control signal. Each one flags a different failure mode \u2014 segmentation errors, ambient contamination, cellular stress, background noise \u2014 and learning to read them individually is what makes joint interpretation possible later. <\/p>\n\n\n\n Very small segmented regions frequently represent debris or incomplete segmentation.<\/p>\n\n\n\n Very large regions often indicate merged neighboring cells (doublets).<\/p>\n\n\n\n Examining the distribution of cell areas helps identify segmentation errors while preserving naturally large cell types.<\/p>\n\n\n\n Transcript density normalizes transcript counts by segmented cell area.<\/p>\n\n\n\n Unlike total transcript count, density distinguishes genuinely transcriptionally active cells from large, poorly segmented regions containing mostly background signals. Low-density objects frequently correspond to ambient RNA contamination or segmentation artifacts, making this one of the most informative QC metrics for imaging-based spatial assays.<\/p>\n\n\n\n Rather than applying a universal cutoff, mitochondrial expression should be evaluated relative to the tissue under study.<\/p>\n\n\n\n Visualizing mitochondrial percentage together with clustering often reveals whether high-mitochondrial cells represent stressed populations or biologically distinct cell types that should be retained.<\/p>\n\n\n\n Negative-control probes estimate the experimental noise floor.<\/p>\n\n\n\n Cells with elevated negative-control signal should be examined alongside transcript density and total transcript counts before deciding whether they represent technical artifacts.<\/p>\n\n\n\n Together, these four metrics form a diagnostic framework for spatial data: area identifies segmentation errors, density distinguishes true transcriptional signal from background, mitochondrial content indicates whether cells are metabolically compromised or represent a distinct biological population, and negative-control probes quantify the baseline rate of non-specific signal. <\/p>\n\n\n\n Interpreting them jointly \u2014 rather than any one alone \u2014 is what allows QC to reflect the actual condition of the tissue, rather than an arbitrary cutoff. <\/p>\n\n\n\n High-quality QC improves every downstream analysis. Removing technical artifacts leads to more stable clustering, clearer cell-type boundaries, more reliable differential expression, and more accurate spatial neighborhood and ligand-receptor analyses. Importantly, thoughtful QC reduces false biological discoveries while preserving rare but meaningful cellular populations.<\/p>\n\n\n\n<\/figure>\n\n\n\n2. Segmentation Artifacts<\/strong><\/h3>\n\n\n\n
3. Mitochondrial Transcripts<\/strong><\/h3>\n\n\n\n
4. Negative Control Probes<\/strong><\/h3>\n\n\n\n
Building Data-Driven QC Thresholds<\/strong><\/h2>\n\n\n\n
<\/figure>\n\n\n\nInteractive Threshold Optimization in SpatialX<\/strong><\/h2>\n\n\n\n
\n
Interpreting Common QC Metrics<\/strong><\/h2>\n\n\n\n
1. Cell Area<\/strong><\/h3>\n\n\n\n
2. Transcript Density<\/strong><\/h3>\n\n\n\n
3. Mitochondrial Percentage<\/strong><\/h3>\n\n\n\n
4. Negative-Control Signal<\/strong><\/h3>\n\n\n\n
From Quality Control to Biological Discovery<\/strong><\/h2>\n\n\n\n
![\"\"](\"https:\/\/bioturing.com\/blog\/wp-content\/uploads\/2026\/09\/image-1-1024x721.png\")
<\/figure>\n\n\n\n![\"\"](\"https:\/\/bioturing.com\/blog\/wp-content\/uploads\/2026\/09\/image-1024x678.png\")
<\/figure>\n\n\n\n