Yila
All use cases
Data Analysis

Single-cell RNA-seq: QC and annotation of 11,769 PBMCs

Upload the PBMC10K QC and marker-expression tables and Yila reads the nCount, nFeature, and mitochondrial-fraction distributions before setting any thresholds, annotates each cluster with a cell-type / positive-marker / lookalikes-to-exclude evidence table, and generates the marker dot plot — preliminary labels are not final biological annotation.

This is QC and marker-expression data for 11,769 PBMC single cells. First check the distributions of nCount_RNA, nFeature_RNA, and percent_mitochondrial, and tell me which cells need a closer look. Do not annotate cell types yet.

Distributions plotted: nFeature peaks near 2,000 with a secondary bump, nCount is right-skewed with a long tail, and mitochondrial fraction concentrates low but tails out — thresholds must follow this dataset, not convention.

Distributions plotted: nFeature peaks near 2,000 with a secondary bump, nCount is right-skewed with a long tail, and mitochondrial fraction concentrates low but tails out — thresholds must follow this dataset, not convention.
Do not apply fixed thresholds blindly. Given this dataset’s distributions, propose an initial QC scheme for teaching purposes, and explain what low nFeature, high nFeature, high nCount, and high mitochondrial fraction may each indicate.

The initial scheme: low nFeature suggests empty droplets or debris, high nFeature and high nCount suggest doublets, and high mitochondrial fraction suggests damaged cells. Cells affected by each reason are counted separately, with notes on what hard thresholds alone cannot decide.

Generate the marker dot plot from this summary table. Dot size is the fraction of cells expressing the marker, color is mean log1p expression, and explain why each cluster is annotated the way it is.

The dot plot is ready: CD3D/IL7R mark T cells, MS4A1/CD79A mark B cells, CD14/LYZ mark monocytes, GNLY/NKG7 mark NK cells, and PPBP/PF4 mark platelets — every annotation ships with its positive markers and the lookalike types ruled out, never a single-marker call.

The dot plot is ready: CD3D/IL7R mark T cells, MS4A1/CD79A mark B cells, CD14/LYZ mark monocytes, GNLY/NKG7 mark NK cells, and PPBP/PF4 mark platelets — every annotation ships with its positive markers and the lookalike types ruled out, never a single-marker call.
Organize the results into three tables: QC thresholds with rationale, cell counts per QC reason, and cell types with marker evidence. State clearly that the thresholds apply only to this dataset and that preliminary marker labels are not final biological annotation.

All three tables are delivered with a QA note: the thresholds hold only for this batch; preliminary marker labels are not final biological annotation; clusters with weak evidence or possible mixtures are listed under “needs review”, with suggestions for extra markers or doublet detection.

pbmc_qc_distributions.pngQC distributions · PNG
pbmc_marker_dotplot.pngMarker dot plot · PNG
qc_threshold_table.csvQC thresholds · CSV
cell_type_marker_evidence.csvAnnotation evidence · CSV
marker_dotplot_script.pyPlotting script · Python
Task completed

Yila task replay complete.

Start with this method