Zhou Y et al., Cell Genomics - This paper introduces scPrediXcan, which combines a deep-learning model (ctPred) built on Enformer-derived features with single-cell RNA-seq to perform cell-type-specific TWAS via a linearized SNP predictor (ℓ-ctPred), improving gene discovery for T2D and SLE. Key terms: cell-type-specific expression, deep learning, TWAS, single-cell RNA-seq, GWAS.

Study Highlights:
The authors developed ctPred, a compact deep-learning model that uses Enformer epigenomic features to predict cell-type-specific pseudobulk expression. They linearized ctPred into ℓ-ctPred (SNP-based elastic net) so predictions can be used with GWAS summary statistics in S-PrediXcan. Applied to type 2 diabetes and systemic lupus erythematosus, scPrediXcan identifies more candidate causal genes, explains more GWAS loci, and provides cell-type-resolved insights compared with canonical TWAS approaches. Models and weights for 46 cell types are released via predictdb.org for community use.

Conclusion:
scPrediXcan leverages sequence-based deep learning and single-cell data to enable cell-type-level TWAS with higher power and broader gene coverage than canonical methods, offering improved mechanistic resolution for complex traits while noting limitations related to model biases and cis-focused inference.

Music:
Enjoy the music based on this article at the end of the episode.

Article title:
scPrediXcan integrates deep learning methods and single-cell data into a cell-type-specific transcriptome-wide association study framework

First author:
Zhou Y

Journal:
Cell Genomics

DOI:
10.1016/j.xgen.2025.100875

Reference:
Zhou Y., Adeluwa T., Zhu L., Salazar-Magaña S., Sumner S., Kim H., Gona S., Nyasimi F., Kulkarni R., Powell J.E., Madduri R., Liu B., Chen M., Im H.K. scPrediXcan integrates deep learning methods and single-cell data into a cell-type-specific transcriptome-wide association study framework. Cell Genomics. 2025;5:100875. doi:10.1016/j.xgen.2025.100875

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 – Stripe donations: https://donate.stripe.com/7sY4gz71B2sN3RWac5gEg00

Official website https://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/scpredixcan-cell-type-twas

QC:
This episode was checked against the original article PDF and publication metadata for the episode release published on 2025-05-22.

QC Scope:
- article metadata and core scientific claims from the narration
- excludes analogies, intro/outro, and music
- transcript coverage: Audited the transcript for substantive scientific content: ctPred architecture, Enformer-derived features, ℓ-ctPred linearization, S-PrediXcan integration, disease results (T2D and SLE) with cell-type specificity, limitations (directionality, distal enhancers, ancestry), and public resources.
- transcript topics: Cell-type-specific TWAS rationale vs bulk tissue TWAS; ctPred architecture and Enformer-derived epigenomic features; ℓ-ctPred linearization and S-PrediXcan integration; Comparison with PEN and scalability considerations; T2D results in islet cell types (gamma, stellate cells) and bulk vs pseudobulk; SLE results in immune cell types (T cells, monocytes; CFB, CXCR5)

QC Summary:
- factual score: 10/10
- metadata score: 10/10
- supported core claims: 8
- 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:
- ctPred uses Enformer-derived epigenomic features and a lightweight four-layer MLP to predict cell-type-specific expression
- ℓ-ctPred is a SNP-based elastic net linearization enabling TWAS with GWAS summary statistics (S-PrediXcan)
- ctPred predictions are cross-validated across cell types with high Pearson correlations (0.753–0.892 reported in the article)
- Enformer input window around TSS is 196,608 bp and yields 5,313 epigenomic features
- T2D: scPrediXcan identified 222 candidate genes across 108 LD blocks vs 12 genes across 11 LD blocks (TWAS-pseudobulk) and 111 genes across 64 LD blocks (TWAS-bulk)
- SLE: scPrediXcan identified 129 genes across 24 LD blocks vs 11 (TWAS-pseudobulk) and 54 (TWAS-bulk) across LD blocks

QC result: Pass.