SUN for scRNA-seq: stable-interval and embedding-based approximate silhouette guided Louvain resolution selection

In single-cell RNA-seq analysis, clustering resolution is often chosen by trial and error, yet it changes reported groups and downstream interpretation. Over-splitting or over-merging reduces reproducibility across teams. A transparent, practical approach to resolution choice is needed, including when over-clustering must be corrected before interpretation. We introduce SUN (Stable-interval guided Unsupervised Neighborhood resolution selection), an unsupervised layer that selects one Louvain-style resolution on a fixed shared nearest-neighbour (SNN) graph. SUN scans a resolution grid with caching, identifies stable intervals as maximal runs of constant cluster count along cached resolutions (plateaus), and ranks intervals using a composite score of embedding-based approximate silhouette and interval width. We evaluated SUN on simulations, eight real scRNA-seq benchmarks, and a cortex over-clustering case. In simulation ( n = 72 conditions), mean SUN ARI was 0.860 ± 0.230 (mean ± SD). Across eight benchmarks, dataset-level SUN achieved mean ± SD ARI 0.864 ± 0.099, FMI 0.886 ± 0.083, and NMI 0.873 ± 0.087. In the cortex over-clustering case study, SUN achieved ARI = 0.996 relative to the Manual Adjust partition used in the scSHC literature while recovering the expected UBC/MLI1 merge direction. Agreement used ARI, Fowlkes–Mallows index, and normalised mutual information, with internal indices as complementary evidence. SUN converts undocumented trial-and-error on the Louvain resolution axis into an explicit, auditable recommendation—stable intervals plus quantitative ranking—within standard Seurat-style workflows. Across the included benchmarks, SUN showed generally high concordance with curated reference annotations, while remaining competitive with alternative unsupervised partitions; a cortex case study of over-clustering correction showed high agreement with the scSHC-associated Manual Adjust evaluation target. Seed sensitivity on some graphs and the heuristic nature of the ranking score are reported as practical caveats rather than fully eliminated sources of variability. An online trial deployment is currently accessible at http://8.166.114.84:3838/SUN_shiny_app_project/ . The SUN codebase is available at https://github.com/BioinfoCenterSYSMH/SUN-Project .

Authors

Institutions

Publication Details

Journal
BMC Bioinformatics
Published
2026-11-14
DOI
https://doi.org/10.1186/s12859-026-06766-9
Primary Topic
Single-cell and spatial transcriptomics
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

SUN for scRNA-seq: stable-interval and embedding-based approximate silhouette guided Louvain resolution selection

Jien Wang, Ge Li, Yiqing Zheng, Jiahao Liu et al.
BMC Bioinformatics
Single-cell and spatial transcriptomics
article

SUN for scRNA-seq: stable-interval and embedding-based approximate silhouette guided Louvain resolution selection

Jien Wang, Ge Li, Yiqing Zheng, Jiahao Liu, Qiaoling Liang, Jianyou Liao, Junbo Wang, Jiaqi Pang, Bing Yang, Maojin Liang
article en

Abstract

In single-cell RNA-seq analysis, clustering resolution is often chosen by trial and error, yet it changes reported groups and downstream interpretation. Over-splitting or over-merging reduces reproducibility across teams. A transparent, practical approach to resolution choice is needed, including when over-clustering must be corrected before interpretation. We introduce SUN (Stable-interval guided Unsupervised Neighborhood resolution selection), an unsupervised layer that selects one Louvain-style resolution on a fixed shared nearest-neighbour (SNN) graph. SUN scans a resolution grid with caching, identifies stable intervals as maximal runs of constant cluster count along cached resolutions (plateaus), and ranks intervals using a composite score of embedding-based approximate silhouette and interval width. We evaluated SUN on simulations, eight real scRNA-seq benchmarks, and a cortex over-clustering case. In simulation ( n = 72 conditions), mean SUN ARI was 0.860 ± 0.230 (mean ± SD). Across eight benchmarks, dataset-level SUN achieved mean ± SD ARI 0.864 ± 0.099, FMI 0.886 ± 0.083, and NMI 0.873 ± 0.087. In the cortex over-clustering case study, SUN achieved ARI = 0.996 relative to the Manual Adjust partition used in the scSHC literature while recovering the expected UBC/MLI1 merge direction. Agreement used ARI, Fowlkes–Mallows index, and normalised mutual information, with internal indices as complementary evidence. SUN converts undocumented trial-and-error on the Louvain resolution axis into an explicit, auditable recommendation—stable intervals plus quantitative ranking—within standard Seurat-style workflows. Across the included benchmarks, SUN showed generally high concordance with curated reference annotations, while remaining competitive with alternative unsupervised partitions; a cortex case study of over-clustering correction showed high agreement with the scSHC-associated Manual Adjust evaluation target. Seed sensitivity on some graphs and the heuristic nature of the ranking score are reported as practical caveats rather than fully eliminated sources of variability. An online trial deployment is currently accessible at http://8.166.114.84:3838/SUN_shiny_app_project/ . The SUN codebase is available at https://github.com/BioinfoCenterSYSMH/SUN-Project .

BMC Bioinformatics
Sun Yat-sen University (CN), Sun Yat-sen Memorial Hospital (CN)
Single-cell and spatial transcriptomics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon