{
  "abstract": "Rare malignant cells in single-cell RNA sequencing (scRNA-seq), including circulating tumor cells (CTCs), minimal residual disease (MRD) cells, and small malignant populations in heterogeneous tumor microenvironments (TME), represent an important clinical challenge. These cells occur at very low frequency, requiring separation from a large non-malignant background. Existing methods show inconsistent evaluation, using different datasets, label definitions, prevalence settings, scoring strategies, and failure handling. We present Rare-cell Evaluation Across Cancer Heterogeneity (REACH), treating inter-lineage rare malignant cell detection as an imbalanced ranking task. REACH benchmarks CNV-supported malignant cells (P_HC) against immune, stromal, and normal epithelial cells (B_HC), modeling the CTC and MRD screening setting. REACH excludes intra-lineage state discovery, including drug-tolerant persisters, epithelial–mesenchymal transition (EMT), and transitional malignant states, evaluating these separately in Track F. REACH combines confidence tiered multi-arm labels with five evaluation tracks, covering controlled real spike-ins, synthetic Splatter stress tests, null background controls, natural blood/CTC prevalence, and supervised label noise diagnostics. Applying standardized method wrappers, fallback-aware metrics, non-parametric statistics, and audited figure regeneration, REACH provides a reproducible evaluation framework. REACH evaluates 10 computational methods on 10 curated scRNA-seq cancer datasets containing 359,739 cells. This benchmark produces 11,250 performance measurements across 1,125 benchmark tasks. The supervised highly variable gene logistic regression ceiling achieves a median average precision of 1.0, indicating strong expression differences between tumor and non-tumor cells. Current unsupervised methods recover only part of this signal. CaSee, serving as a cancer/normal cell scorer rather than a rare-cell detector, reaches a median average precision of 0.512 (0.606 excluding 20 degenerate units). FiRE, the strongest published inter-lineage detector, achieves a median average precision of 0.313. Statistical testing confirms significant performance differences among methods. REACH shows strong context dependence, with no method performing best across all tumor types. Track F further shows performance depending strongly on the inter-lineage malignant-versus-microenvironment contrast. Restricting the background to normal epithelial cells of the same lineage, nine of ten methods fall to near-chance average precision. These results indicate that inter-lineage leaderboards should not be extrapolated to intra-lineage rare-cell discovery.",
  "abstract_shown": true,
  "affiliations": [
    "Department of Computer Science and Engineering, School of Engineering and Sciences, SRM University-AP , , ,",
    "Artificial Intelligence Technology Institute, SRM University-AP , , ,"
  ],
  "article_type": "research article",
  "arxiv": null,
  "author_total": 5,
  "authors": [
    {
      "affs": [
        0
      ],
      "name": "Moram Venkata Satya Jaswanth"
    },
    {
      "affs": [
        0
      ],
      "name": "Amruthaluri Gavin"
    },
    {
      "affs": [
        0
      ],
      "name": "Padarthi Aakash"
    },
    {
      "affs": [
        0
      ],
      "name": "Muttha Tejesh Kumar"
    },
    {
      "affs": [
        1
      ],
      "name": "Abhijit Dasgupta"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/1hs7ecz42djq1.bib",
  "cited_by": 0,
  "doi": "10.3389/fbinf.2026.1878538",
  "fulltext_url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13724523/?tool=EBI",
  "id": "1hs7ecz42djq1",
  "issns": [
    "2673-7647"
  ],
  "landing_url": "https://europepmc.org/article/PMC/PMC13724523",
  "language": "eng",
  "licence": "CC-BY",
  "licence_label": "CC BY",
  "licence_source": "article",
  "licence_url": "https://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": null,
  "native_subjects": [
    [
      "heading",
      "Original Research"
    ],
    [
      "openalex",
      "1312|Single-cell and spatial transcriptomics"
    ]
  ],
  "openalex": "W7222978329",
  "pdf_url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13724523/pdf/?tool=EBI",
  "pmcid": "PMC13724523",
  "pmid": "42973741",
  "published": "2026-10-19",
  "publisher": "Frontiers Media SA",
  "ris": "https://research.draconitic.com/paper/1hs7ecz42djq1.ris",
  "sources": [
    {
      "first_seen": "2026-11-03T15:20:51Z",
      "native_id": "PMC:PMC13724523",
      "source": "europepmc",
      "url": "https://europepmc.org/article/PMC/PMC13724523"
    },
    {
      "first_seen": "2026-11-04T12:23:05Z",
      "native_id": "PMC13724523",
      "source": "pmc",
      "url": "https://doi.org/10.3389/fbinf.2026.1878538"
    },
    {
      "first_seen": "2026-11-04T15:20:48Z",
      "native_id": "MED:42973741",
      "source": "europepmc",
      "url": "https://europepmc.org/article/MED/42973741"
    }
  ],
  "subjects": [
    {
      "field": "biochemistry",
      "origin": "openalex:1312|Single-cell and spatial transcriptomics",
      "subfield": "molecular-biology"
    }
  ],
  "title": "REACH: a reproducible benchmark for inter-lineage rare malignant cell detection in single-cell RNA sequencing",
  "title_text": "REACH: a reproducible benchmark for inter-lineage rare malignant cell detection in single-cell RNA sequencing",
  "type": "peer-reviewed",
  "updated_at": "2026-11-06T12:20:58Z",
  "url": "https://research.draconitic.com/paper/1hs7ecz42djq1",
  "venue": "Frontiers in bioinformatics",
  "versions": []
}