Bio Single Cell Splicing

Version Compatibility

Reference examples tested with: MARVEL 2.0+, BRIE2 0.2.4+, scQuint 0.1+, SpliZ 0.0.1+, Sierra 1.0+, Psix 0.1+, anndata 0.10+, scanpy 1.10+, pandas 2.2+, scipy 1.13+

Before using code patterns, verify installed versions match. If versions differ:

• Python: pip show then help(module.function) to check signatures
• R: packageVersion('') then ?function_name to verify parameters
• CLI: --version then --help to confirm flags

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

Single-Cell Splicing Analysis

The fundamental decision is chemistry, not tool. Most droplet 3' scRNA-seq cannot support transcriptome-wide splicing inference because reverse transcription primes from the poly(A) tail and most reads land in the 3' UTR — far from CDS-region splicing events. Plate-based full-length methods and single-cell long-read sequencing are the chemistries that give per-cell isoform structure across the gene body.

The 10X 3' Problem (Quantified)

Three compounding mechanisms make 10X Chromium 3' (v3.1, GEM-X, v4) hostile to splicing:

1. 3' enrichment: median fragment 70% of unique reads fall within 3' UTR.
2. Short R2 (~91 nt): each read straddles at most one junction; usually none, because R2 lands in 3' UTR.
3. PCR concatemers and TSO artifacts: pollute junction detection; UMI collapse is gene-level, not isoform-level.

Quantitative estimate: Only a small fraction of cassette exons sit close enough to the polyA site to be sampled by 3' chemistry (empirical estimates from APA/3'-end atlases — see Tian & Manley 2017 Nat Rev Mol Cell Biol for the 3' UTR isoform landscape). Effective junction read yield from 10X 3' is ** 1.5 and pvalue wide reshape.

BRIE2: TensorFlow Memory

Trigger: Large cohort (>10k cells) with deep coverage.

Mechanism: Variational inference loads full count matrix; TensorFlow allocates GPU memory aggressively.

Symptom: OOM kills; training stalls.

Fix: Reduce --batchSize from default (500000) to 100000 or 50000; train per-chromosome batch; use CPU mode for very small cohorts. Note flag is camelCase --batchSize, not --batch_size.

scQuint: 3' Data Sparsity

Trigger: Running scQuint on 10X 3' v3 data hoping for splicing signal.

Mechanism: scQuint's latent Dirichlet model needs junction counts; 10X 3' yields too few junction reads to fit the model robustly.

Symptom: All cells assign to one cluster; no informative splicing signal.

Fix: Pivot to APA analysis with Sierra; or upgrade chemistry to MAS-Iso-seq.

Psix: Missing kNN Graph

Trigger: Running Psix without precomputed cell-cell graph.

Mechanism: Psix tests PSI smoothness on a pre-existing cell-cell graph; without one, no smoothness statistic.

Symptom: Empty results or error about missing connectivities.

Fix: Run sc.pp.neighbors(adata) before Psix; ensure connectivities is in adata.obsp.

Sierra: Annotation Gaps

Trigger: GTF missing 3'UTR annotations.

Mechanism: Sierra peak-calls within annotated 3'UTRs; missing annotations mean missed peaks.

Symptom: Few peaks detected; gene-level coverage but no APA calls.

Fix: Use comprehensive GENCODE annotation; or run de-novo peak calling first.

Reconciliation: When Single-Cell Tools Disagree

| Pattern | Likely cause | Action | |---------|--------------|--------| | MARVEL sig, BRIE2 not | Per-cell PSI noise (BRIE2 conservative); MARVEL pseudobulk-like | Trust MARVEL for cell-type comparisons; BRIE2 for within-cluster | | BRIE2 sig, MARVEL not | Cell-state effect smoother than cell-type boundary | Test along trajectory with Psix | | SpliZ sig, MARVEL not | Annotation-free SpliZ catches novel events | Investigate junction structure manually | | Sierra sig, MARVEL not | Sierra is APA, MARVEL is splicing — different biology | Distinguish in interpretation | | Pseudobulk sig, per-cell not | Power issue; effect averaged out per-cell | Report at cluster level, not per-cell |

Quantitative Concepts Unique to Single-Cell

Per-cell PSI vs pseudobulk PSI:

• Per-cell PSI: meaningful only when junction coverage exceeds ~10-20 reads per cell per event (plate-based or long-read).
• Pseudobulk PSI: aggregate, recovers bulk-level statistical power, discards within-cluster heterogeneity.

Modality detection in PSI distributions (Song 2017 Mol Cell): | Modality | PSI distribution | Biology | |----------|------------------|---------| | Included | Peaked at 1 | Constitutive inclusion | | Excluded | Peaked at 0 | Constitutive skipping | | Bimodal | Mixture at 0 and 1 | Mixed cell states or monoallelic-like bursting | | Middle | Peaked ~0.5 | Often technical (well-contamination, doublets, or low-coverage shrinkage to prior); confirm with full-length | | Multimodal | Multiple peaks | Complex regulation; deserves follow-up |

Beta-binomial vs binomial models: with sparse counts, binomial PSI is overdispersed. Beta-binomial models (BRIE2; leafcutter2 as Dirichlet-multinomial cluster-level) handle this. For very sparse droplet data, even beta-binomial fits poorly per cell — collapse to pseudobulk.

Imputation pitfalls: naive imputation (MAGIC, scImpute, ALRA) of expression matrices is not appropriate for PSI: imputing missing junction counts averages over neighboring cells and obliterates the very heterogeneity under study. Psix's approach — testing smoothness of observed PSI on the kNN graph — is the principled alternative.

Cell-Type-Specific Splicing Biology

| System | Event | Regulator | |--------|-------|-----------| | Neural microexons | 3-27 nt exons enriched in brain | SRRM4 (nSR100); SRRM3 in retina (Irimia 2014 Cell) | | Neural differentiation | PTBP1 -> PTBP2 switch | miR-124 represses PTBP1; derepresses neural exons (Boutz 2007 Genes Dev) | | T-cell activation | CD45 RA -> RO | hnRNP-L, ESRP-mediated | | Erythropoiesis | EPB41 exon 16 | Splicing factor switching during maturation | | Cardiac development | TTN N2BA -> N2B | MBNL1/CELF1 antagonism | | EMT | FGFR2 IIIb -> IIIc, ENAH exon 11a | ESRP1/2 loss in mesenchymal state (Warzecha 2009 Mol Cell) | | Activated T cell | CD45 isoform shift | Multiple SR/hnRNP regulators |

Quality Thresholds

| Metric | Recommendation | |--------|----------------| | Cells per event with reads | >=50 (per-cell PSI); >=200 cells per cluster (pseudobulk) | | Junction reads per event per cell | >=5 with coverage; 0.2 between clusters | | Library | full-length plate or long-read for transcriptome-wide; 3' for APA only | | Doublet filtering | Required before splicing analysis (DoubletFinder, Scrublet) | | Cells per cluster (pseudobulk) | >=100 ideal; >=50 minimum | | nuclear vs whole-cell | snRNA-seq enriches IR; treat with caution |

Common Errors

| Error | Cause | Solution | |-------|-------|----------| | MARVEL: ComputePSI returns empty | STAR SJ.out.tab missing strand info | Re-run STAR with --outSJtype Standard | | brie.tl.fit: NaN loss | Insufficient junction reads per cell | Filter cells with min_reads=20; raise threshold | | scQuint: convergence not reached | LDA model fit on too-few junctions | Aggregate by chromosome; or switch chemistry | | Psix: missing connectivities | Neighbors graph not computed | Run sc.pp.neighbors(adata) first | | Sierra: no peaks called | GTF missing 3'UTR annotations | Use comprehensive GENCODE; or de-novo peak-call | | MARVEL: ggplot error | Seurat version mismatch | Match MARVEL and Seurat versions | | FLAMES: barcode rescue failed | Short-read 10X output not in expected directory | Verify cellranger output structure |

Common Pitfalls

• Treating 10X 3' splicing analysis as legitimate — the chemistry doesn't support it. Use Sierra for APA or upgrade to MAS-Iso-seq.
• Imputing PSI matrices — destroys the heterogeneity to be detected. Use Psix or BRIE2 instead.
• Per-cell PSI on droplet data — typically too sparse for stable estimates. Use pseudobulk first, then drill down to per-cell.
• Confusing APA with splicing — Sierra results look like AS but are 3' UTR isoforms. Different machinery, different biology.
• snRNA-seq IR signal misinterpreted as splicing dysregulation — nuclear RNA is enriched for incompletely spliced transcripts; baseline IR is high.
• Trusting per-cell PSI from BRIE2 without ELBOgain test — BRIE2's per-cell point estimates are noisy; the principled output is the ELBOgain cell-state-association statistic.
• Microexon analysis with default short-read aligners — anchors >=20 nt miss most microexons; use VAST-TOOLS, MicroExonator, or long-read.
• Skipping doublet filtering before splicing — doublets create artificial PSI mid-modality.

Related Skills

• single-cell/preprocessing - QC and normalization (must run before splicing)
• single-cell/clustering - Cell type annotation prerequisite
• single-cell/doublet-detection - Doublet filtering critical for splicing
• single-cell/data-io - h5ad / Seurat I/O
• splicing-quantification - Bulk RNA-seq comparison context
• long-read-splicing - Full-isoform analysis from MAS-Iso-seq, scISOr-Seq2; future of single-cell splicing

References

• Huang & Sanguinetti 2021 Genome Biol - BRIE2
• Wen et al 2023 Nucleic Acids Research 51:e29 - MARVEL
• Benegas, Fischer & Song 2022 eLife - scQuint (annotation-free single-cell splicing analysis, validated on Smart-seq2)
• Olivieri et al 2022 Nat Methods - SpliZ
• Buen Abad Najar et al 2022 Genome Research 32:1385 - Psix
• Patrick et al 2020 Genome Biol - Sierra
• Song et al 2017 Mol Cell - splicing modality classification
• Picelli et al 2014 Nat Protoc - Smart-seq2
• Hagemann-Jensen et al 2020 Nat Biotech - Smart-seq3
• Hagemann-Jensen et al 2022 Nat Biotech - Smart-seq3xpress
• Hahaut et al 2022 Nat Biotech - FLASH-seq
• Salmen et al 2022 Nat Biotech - VASA-seq
• Johnson et al 2023 Nat Commun - STORM-seq
• Al'Khafaji et al 2024 Nat Biotech - MAS-Iso-seq / Kinnex
• Tian et al 2021 Genome Biology 22:310 - FLAMES
• Joglekar et al - scISOr-Seq2 mouse cortex atlas (consult most recent publication for venue/year)
• Irimia et al 2014 Cell - neural microexons / SRRM4
• Boutz et al 2007 Genes Dev - PTBP1/PTBP2 neural switch
• Tian & Manley 2017 Nat Rev Mol Cell Biol - alternative polyadenylation and 3' UTR isoforms

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: GPTomics
• Source: GPTomics/bioSkills
• License: MIT

Install and usage instructions live in the source repository linked above.