scCertify

Explainable Confidence Scoring for Single-Cell RNA-seq Annotations

scCertify is an R package for evaluating the reliability of single-cell RNA-seq annotations using explainable confidence scoring.

The framework integrates:

• UCell-based marker enrichment
• Neighborhood agreement scoring
• Entropy-based uncertainty estimation
• Doublet-aware confidence modeling
• Ontology-aware label matching
• Confidence calibration
• Explainable confidence attribution

scCertify aims to provide a biologically interpretable framework for identifying reliable and uncertain cell annotations in single-cell datasets.

---

Why scCertify?

Most annotation tools assign labels without estimating how trustworthy those labels are.

In real single-cell datasets, uncertainty can arise from:

• Transitional cellular states
• Technical doublets
• Sparse transcriptomic profiles
• Weak marker enrichment
• Reference atlas mismatch
• Ambiguous neighborhood structure
• Batch effects

scCertify quantifies annotation reliability and explains why cells are considered uncertain.

---

Features

Current Features

• Confidence scoring for single-cell annotations
• UCell-based marker enrichment scoring
• kNN neighborhood agreement analysis
• Entropy-based uncertainty estimation
• Doublet-aware confidence scoring
• Ontology-aware label matching
• Confidence calibration
• Confidence classification
• Explainable confidence attribution
• Seurat integration
• Publication-ready visualization support

---

Workflow Overview

Single-cell RNA-seq Data
            ↓
      SingleR Annotation
            ↓
     Marker Enrichment
         (UCell)
            ↓
   Neighborhood Agreement
            ↓
    Entropy Uncertainty
            ↓
     Doublet Detection
      (scDblFinder)
            ↓
   Confidence Calibration
            ↓
 Explainable Confidence
            ↓
 Final Confidence Score
   + Confidence Class

---

Installation

Install Dependencies

install.packages(c(
  "Seurat",
  "ggplot2",
  "FNN",
  "entropy",
  "devtools"
))

install.packages("BiocManager")

BiocManager::install(c(
  "SingleR",
  "celldex",
  "SingleCellExperiment",
  "UCell",
  "scDblFinder"
))

---

Install scCertify

devtools::install_github(
  "Jaya-Surya-dev/scCertify"
)

---

Quick Start

Load Libraries

library(scCertify)

library(Seurat)

library(SingleR)

library(celldex)

library(UCell)

library(scDblFinder)

---

Load Example Dataset

data("pbmc_small")

---

Preprocess Data

pbmc_small <- NormalizeData(pbmc_small)

pbmc_small <- FindVariableFeatures(pbmc_small)

pbmc_small <- ScaleData(pbmc_small)

pbmc_small <- RunPCA(pbmc_small)

pbmc_small <- RunUMAP(
  pbmc_small,
  dims = 1:10
)

---

Run SingleR Annotation

sce <- as.SingleCellExperiment(
  pbmc_small
)

ref <- HumanPrimaryCellAtlasData()

pred <- SingleR(
  test = sce,
  ref = ref,
  labels = ref$label.main
)

pbmc_small$predicted_label <- pred$labels

---

Detect Doublets

sce <- scDblFinder(sce)

pbmc_small$doublet_score <-
  colData(sce)$scDblFinder.score

pbmc_small$doublet_class <-
  colData(sce)$scDblFinder.class

---

Define Marker Database

markers <- list(

  "B_cell" = c(
    "MS4A1",
    "CD79A"
  ),

  "T_cells" = c(
    "CD3D",
    "IL7R"
  ),

  "Monocyte" = c(
    "LYZ",
    "S100A8"
  ),

  "NK_cell" = c(
    "NKG7",
    "GNLY"
  ),

  "DC" = c(
    "FCER1A",
    "CST3"
  ),

  "Platelets" = c(
    "PPBP",
    "PF4"
  )
)

---

Calculate Entropy

pbmc_small$entropy_score <-
  entropy_score(
    pred$scores
  )

pbmc_small$entropy_norm <- (

  pbmc_small$entropy_score -

  min(pbmc_small$entropy_score)

) / (

  max(pbmc_small$entropy_score) -

  min(pbmc_small$entropy_score)

)

---

Run scCertify

pbmc_small <- cell_certify(
  pbmc_small,
  markers
)

---

Example Outputs

Confidence Score UMAP

FeaturePlot(
  pbmc_small,
  features = "confidence_score"
)

---

Confidence Classes

DimPlot(
  pbmc_small,
  group.by = "confidence_class"
)

---

Entropy Landscape

FeaturePlot(
  pbmc_small,
  features = "entropy_norm"
)

---

Explain Confidence Attribution

cell_id <- colnames(
  pbmc_small
)[1]

explain_confidence(
  pbmc_small,
  cell_id
)

Example output:

[1] "Weak marker enrichment"
[2] "High annotation uncertainty"
[3] "Possible doublet contamination"

---

Confidence Framework

The current confidence model integrates:

• Marker enrichment
• Neighborhood agreement
• Entropy certainty
• Doublet probability

Current scoring framework:

Confidence = 0.35(Marker Score) + 0.35(Neighborhood Agreement) + 0.20(Entropy Certainty) - 0.10(Doublet Probability)

---

Package Structure

scCertify/

├── R/
│   ├── calibrate_confidence.R
│   ├── cell_certify.R
│   ├── classify_confidence.R
│   ├── entropy_score.R
│   ├── explain_cell.R
│   ├── explain_confidence.R
│   ├── marker_score.R
│   ├── match_labels.R
│   └── neighbor_score.R
│
├── man/
├── DESCRIPTION
├── NAMESPACE
├── README.md
└── LICENSE

---

Planned Features

• Trajectory-aware confidence scoring
• Multimodal confidence integration
• Spatial transcriptomics support
• Automatic marker retrieval
• Cell ontology integration
• Batch-aware confidence estimation
• Calibration models
• Benchmarking framework
• Explainable AI visualization
• Atlas-scale optimization

---

Citation

If you use scCertify in your work, please cite:

Doddetipalli JS.
scCertify: Explainable confidence scoring for
single-cell RNA-seq annotations.

---

Author

Jaya Surya Doddetipalli

---

License

MIT License