Single-cell QC (mitochondrial filtering)

Single-cell QC decides which barcodes/nuclei enter analysis—classic filters (including high mitochondrial content) remove damaged cells but can also discard biologically real stressed or metabolically altered states.

Pipelines look objective. Filtering choices change who is in the atlas—especially for disease or perturbation studies—and foundation models inherit those choices.

Claim ledger

What the evidence shows

Drawn from 3 studies in this library. Mix labels say which citation roles are present; they are not a strength score. Supports means evidence for a finding; Challenges means evidence against a stated position; Qualifies marks scope.

• ○ Limited citations

  High-mito filters can deplete viable metabolically altered malignant cells.

  Filtering cells with high mitochondrial content is shown to remove not only debris but also viable metabolically altered malignant cells—QC as biology, not just cleanup.

  1 study
  1. 1High-mito filters can discard living cancer cell states

• ○ Limited citations

  Foundation cell models for post-perturbation prediction need careful benchmarks.

  Benchmarking foundation cell models for post-perturbation RNA-seq prediction tests how well learned representations generalize after QC and perturbation—downstream of filtering choices.

  1 supporting · 1 qualifying

  Supports

  1. 1Simple baselines beat cell foundation models on Perturb-seq

  Qualifies

  1. 1High-mito filters can discard living cancer cell states— model benchmark ≠ mito-filter depletion experiment

  Study comparison

  Study	Role	Design	N	Population	Outcome
  Simple baselines beat cell foundation models on Perturb-seq2025	Supports	Computational / modellingBenchmark of scGPT and scFoundation against train-mean and classical ML on public Perturb-seq splits	Adamson 68,603 cells, Norman 91,205 cells, plus Replogle subsets; 16,239 genes	Public CRISPR perturbation scRNA-seq datasets (CRISPRi/a screens)	Train-mean and feature-based classical ML outperform scGPT/scFoundation on post-perturbation expression prediction under current benchmarks
  High-mito filters can discard living cancer cell states2025	Qualifiesmodel benchmark ≠ mito-filter depletion experiment	Computational / modellingReanalysis of public cancer scRNA-seq and spatial datasets comparing pctMT filters, viability proxies, metabolism, and clinical correlates	N=134 · Nine public scRNA-seq datasets: 441,445 cells from 134 patients (160,225 malignant)	Malignant and tumour-microenvironment cells across multiple human cancer types	Standard >15% pctMT filters deplete viable, metabolically altered malignant HighMT cells linked to xenobiotic metabolism, drug resistance, and clinical features

• ○ Limited citations

  Disease atlases depend on QC’d nuclei before cell-state claims.

  ICM snRNA-seq profiles ~100k QC’d cardiac nuclei before composition claims—reminding students that atlas biology sits atop filtering decisions.

  1 supporting · 1 qualifying

  Supports

  1. 1What cell programs mark ischemic cardiomyopathy?

  Qualifies

  1. 1High-mito filters can discard living cancer cell states— heart atlas composition ≠ mito-filter methods paper

  Study comparison

  Study	Role	Design	N	Population	Outcome
  What cell programs mark ischemic cardiomyopathy?2023	Supports	Case-controlsnRNA-seq of non-infarct LV from ICM transplant recipients vs NF controls	N=15 · 7 ICM + 8 NF; 99,684 nuclei after QC	Human LV nuclei from ICM and non-failing hearts	Cell-composition and transcriptional differences in ICM vs NF
  High-mito filters can discard living cancer cell states2025	Qualifiesheart atlas composition ≠ mito-filter methods paper	Computational / modellingReanalysis of public cancer scRNA-seq and spatial datasets comparing pctMT filters, viability proxies, metabolism, and clinical correlates	N=134 · Nine public scRNA-seq datasets: 441,445 cells from 134 patients (160,225 malignant)	Malignant and tumour-microenvironment cells across multiple human cancer types	Standard >15% pctMT filters deplete viable, metabolically altered malignant HighMT cells linked to xenobiotic metabolism, drug resistance, and clinical features

Debates

Tensions and limits

Some items are genuine disagreements on the same question. Others mark different assays, populations, or outcomes.

Timeline

How understanding moved

Study years are when the paper was published. Evidence edits are dated changes to this page's claims. Explanations are when PaperFren added a Discovery — not a claim that the science happened that day.

Papers

3 studies in this library bear on Single-cell QC (mitochondrial filtering), ordered by citations.

• What cell programs mark ischemic cardiomyopathy?

  snRNA-seq of ~100k cardiac nuclei showed fewer cardiomyocytes and more specialized endothelial states in ICM, overlapping other end-stage cardiomyopathy programs.

  Cell reports · 2023 · 93 citations

• High-mito filters can discard living cancer cell states

  Across ~441k cells from 134 patients, malignant cells often exceed 15% mitochondrial RNA without stress-death signatures and show metabolic programs relevant to therapy.

  Genome biology · 2025 · 34 citations

• Simple baselines beat cell foundation models on Perturb-seq

  Benchmarking scGPT and scFoundation against baselines finds that training-set mean and basic ML with biological features outperform the foundation models on post-perturbation expression prediction.

  BMC genomics · 2025 · 34 citations

Questions

What is still open

1. A mito-filter methods result, a foundation-model benchmark, and a disease atlas answer different layers of the single-cell stack.

   Watch this

Ask PaperFren about Single-cell QC (mitochondrial filtering)