426: ProtoCloud — Prototypical self-explaining model for single-cell analysis

Guo K et al., Cell Genomics - ProtoCloud is a self-explaining deep generative model that embeds single cells around cell-type-specific prototypes to deliver accurate, uncertainty-aware cell type annotation and gene-level explanations from raw UMI counts. Key terms: single-cell, explainable AI, prototypical models, cell type annotation, uncertainty estimation. Study Highlights:ProtoCloud achieves accurate and efficient annotation of single-cell data, including improved detection of rare cell types, by organizing embeddings around learned prototypes. A disentangled latent space separates biological identity from batch and nuisance variation, improving robustness and label transfer. Built-in uncertainty quantification based on cell–prototype similarity identifies and enables correction of misannotations. Prototypical relevance propagation backpropagates similarity to highlight genes driving classification for instant gene‑level explainability. Conclusion:By combining a decomposed VAE, learnable prototypes, PRP-based gene relevance, and calibrated similarity-based uncertainty, ProtoCloud provides accurate, interpretable, and robust single-cell annotations that detect rare states, correct label errors, and nominate marker genes to support atlas construction and disease studies. Music:Enjoy the music based on this article at the end of the episode. Article title:ProtoCloud: A prototypical self-explaining model for single-cell analysis First author:Guo K Journal:Cell Genomics DOI:10.1016/j.xgen.2026.101217 Reference:Guo K. & Ding J. ProtoCloud: A prototypical self-explaining model for single-cell analysis. Cell Genomics 6, 101217 (2026). doi:10.1016/j.xgen.2026.101217 License:This episode is based on an open-access article published under the Creative Commons Attribution 4.0 International License (CC BY 4.0) – https://creativecommons.org/licenses/by/4.0/ Support:Base by Base is independent and ad-free — no sponsors, no paywall. If an episode was worth your time, chip in and keep the papers audited and the original songs coming:❤️ Support monthly: https://buy.stripe.com/cNifZhclVebvagk2JDgEg01☕ One-time donation: https://donate.stripe.com/7sY4gz71B2sN3RWac5gEg00 More at basebybase.com On PaperCast Base by Base you'll discover the latest in genomics, functional genomics, structural genomics, and proteomics. Episode link: https://basebybase.com/episodes/protocloud-prototypical-self-explaining-single-cell QC:This episode was checked against the original article PDF and publication metadata for the episode release published on 2026-07-23. QC Scope:- article metadata and core scientific claims from the narration- excludes analogies, intro/outro, and music- transcript coverage: Audited the transcript segments describing ProtoCloud architecture, training, uncertainty quantification, and key biological validations (PBMC, RGC time course, EoE).- transcript topics: ProtoCloud architecture and prototypes; Disentangled latent space with z1 and z2; Prototypical relevance propagation (PRP) and HRGs; Robustness to label noise (20% perturbation); PBMC30K annotation corrections (NKG7 example); Time-course retinal ganglion cells after optic nerve crush QC Summary:- factual score: 10/10- metadata score: 10/10- supported core claims: 7- claims flagged for review: 0- metadata checks passed: 4- metadata issues found: 0 Metadata Audited:- article_doi- article_title- article_journal- license Factual Items Audited:- ProtoCloud uses six prototypes per cell type by default- Latent space is partitioned into two components: z1 for cell-type identity and z2 for batch/noise factors- PRP identifies gene-level relevance and HRGs (e.g., CD79B, LY9)...