---
vault_clearance: KETER
halo:
  classification: INTERNAL
  front: "Front MarathonLament"
  created: 2026-03-28
---

# BOUNTY BOARD — MarathonLament (Spliceosome Catalogue)

> **Current Tier:** DEMONSTRATED (3 independent measurements confirm operator hierarchy)
> **Core Question:** How does the spliceosome treat each operator's genes differently, and how does that change during disease?

## SOLVED

| ID | Bounty | Date | Evidence |
|----|--------|------|----------|
| S1 | **Splice entropy differs by operator** — RIBO entropy 0.167, GOLGI 0.388. Cohen's d = 0.79. Breakthrough 14. | 2026-03-28 | SG-NEx nanopore K562 |
| S2 | **GENCODE confirms operator splice potential** — GOLGI 27.8 mean isoforms, RIBO 22.2 (1.25x). RIBO 83% retained intron annotated but SUPPRESSED. GOLGI 3.4x more NMD transcripts. | 2026-03-28 | Ensembl REST API, 410 genes |
| S3 | **All transcripts get more structured in senescence** — Every spotlight gene increases reads/UMI. FTH1 +0.140 (highest). Operator hierarchy: GOLGI > MITO > NUCLEAR > RIBO. | 2026-03-28 | molecule_info.h5, 941M molecules |
| S4 | **RPL5/RPL11 are structurally distinct from other RPLs** — 5% higher reads/UMI, ultra-rigid splice entropy (0.002), stress sensors not structural proteins. | 2026-03-28 | molecule_info.h5 + SG-NEx + literature |
| S5 | **Molecule-level UMI diversity confirms operator predictions** — RIBO diversity -6%, GOLGI reads/UMI +7.3% in senescence. | 2026-03-27 | molecule_info.h5 |

---

## OPEN — STRUCTURAL PREDICTION (ViennaRNA)

| ID | Bounty | Status | Impact |
|----|--------|--------|--------|
| ML1 | **ViennaRNA fold prediction for all operator genes** — MFE structure for RIBO, MITO, GOLGI, NUCLEAR genes. Compare MFE/nt, fraction paired, stem density, G4 potential across operators. Prediction: GOLGI more structured per nucleotide. | IN PROGRESS (agent running) | **CRITICAL** |
| ML2 | **Validate FTH1 IRE prediction** — RNAfold should predict the known IRE stem-loop in FTH1 5'UTR. If it does, method validated. If not, window size or energy model needs adjustment. | IN PROGRESS | **CRITICAL** — method validation |
| ML3 | **Correlate predicted MFE with observed reads/UMI** — For each gene: does more negative MFE predict higher reads/UMI? If r > 0.5, structure predictions explain sequencing behavior. | IN PROGRESS | **CRITICAL** |
| ML4 | **RPL5 structural element near 5S rRNA binding domain** — Does RNAfold predict a structural feature in RPL5 mRNA that corresponds to the domain that binds 5S rRNA? The protein-RNA interaction might be encoded in the mRNA structure. | OPEN | HIGH |
| ML5 | **Chr11 hub pair fold prediction** — What structures do ENSG00000255029 and ENSG00000254526 form? Are they G-quadruplex-rich (COPI binding)? Do they have long-range base pairs (circularization candidates)? | OPEN | **CRITICAL** |

---

## OPEN — BAM COVERAGE SHAPES

| ID | Bounty | Status | Impact |
|----|--------|--------|--------|
| ML6 | **Per-gene coverage profiles P vs S from indexed BAM** — 27 target genes × 5 BAMs. Splice ratio, coverage entropy, 5'/3' bias, truncation point. Running. | IN PROGRESS (task bbx2wgr4r) | **CRITICAL** |
| ML7 | **Coverage holes at predicted stem-loops** — Overlay ViennaRNA fold predictions onto BAM coverage. Do coverage drops occur WHERE stems form? Match = transcript IS folded as predicted. | OPEN (needs ML1 + ML6) | **CRITICAL** — links prediction to observation |
| ML8 | **Retained intron detection from BAM** — For RIBO genes (83% have annotated retained introns): do senescent cells show MORE intronic reads? The "loaded gun" hypothesis — retained introns firing in disease. | OPEN (needs ML6) | HIGH |

---

## OPEN — SPLICEOSOME CATALOGUE DATABASE

| ID | Bounty | Status | Impact |
|----|--------|--------|--------|
| ML9 | **Build the per-operator splice profile table** — For every gene in each operator: (a) GENCODE annotated isoforms, (b) SG-NEx observed entropy, (c) reads/UMI from molecule data, (d) ViennaRNA MFE, (e) BAM coverage shape. One table, all measurements, all operators. | OPEN (needs ML1 + ML6) | **CRITICAL** — the catalogue |
| ML10 | **Entropy-per-potential ratio as the "suppression score"** — Observed entropy / annotated isoforms = how much the spliceosome SUPPRESSES diversity. RIBO = 0.0075 (most suppressed). GOLGI = 0.014. This IS the spliceosome's operator-specific policy. | OPEN | HIGH |
| ML11 | **NMD transcript analysis per operator** — GOLGI has 3.4x more NMD transcripts than RIBO. Are these "experimental" splice forms that get tested and degraded? Or are they functional in specific conditions? Check if NMD transcript expression changes in senescence. | OPEN | HIGH |
| ML12 | **MaxEntScan splice site strength per operator** — Score every splice junction from sequence alone. Do RIBO genes have stronger splice sites (= less alternative splicing) than GOLGI genes? Deterministic, zero parameters, sequence-only. | OPEN | **CRITICAL** — the mechanistic mapping |

---

## OPEN — DISEASE SPLICE ATLAS

| ID | Bounty | Status | Impact |
|----|--------|--------|--------|
| ML13 | **Run spliceosome analysis on Noah WI-38 time course** — Do retained introns increase between day 2 and day 4 (the phase transition)? Does the splice entropy shift PRECEDE or FOLLOW the coupling tensor collapse? | OPEN (needs Noah BAMs from SRA) | **CRITICAL** |
| ML14 | **Run on Replogle Perturb-seq** — When you knock down RPL5 (stress sensor), does the spliceosome behavior of GOLGI genes change? Cross-operator splice effects from CRISPRi. | OPEN (needs Replogle BAMs from SRA) | HIGH |
| ML15 | **Run on LARRY fate data** — The count matrix couldn't distinguish neutrophil vs monocyte fate. Can the SPLICE PATTERN distinguish them? Different isoforms of the same genes in fate-committed progenitors. | OPEN (needs LARRY BAMs from SRA) | **CRITICAL** — the fate prediction rescue |
| ML16 | **SenCat splice analysis** — 14 cell types × 30 senescence paradigms. Does each cell type's splice pattern change differently during senescence? Operator-resolved splice entropy per cell type per senescence method. | OPEN (needs SenCat data from Carlos) | **CRITICAL** |

---

## OPEN — WEIRD TRANSCRIPT REGISTRY

| ID | Bounty | Status | Impact |
|----|--------|--------|--------|
| ML17 | **Catalogue all 250-320 "sequencing-resistant" genes per sample** — High reads/UMI + low expression. Classify by: operator, predicted fold (ViennaRNA), known structural elements, circRNA database overlap. | OPEN (needs ML1) | HIGH |
| ML18 | **Cross-reference with circBase/CIRCpedia** — How many sequencing-resistant genes overlap with known circRNAs? CircRNAs resist sequencing because they have no free 3' end. | OPEN | HIGH |
| ML19 | **HOMER-style motif search in weird transcripts** — For each sequencing-resistant gene: tile the full mRNA into k-mers, search the BAM fragments for enrichment patterns. Truncation, circularization, hairpin signatures. | OPEN (needs ML6) | HIGH |
| ML20 | **The "blackberry phone" catalogue** — Genes whose operator function changes between conditions. mRNA in P, structural RNA in S. Defined by: (a) splice entropy change, (b) reads/UMI change, (c) coverage shape change, (d) predicted fold change. All four measurements must agree. | OPEN (needs all of ML1-ML9) | **CRITICAL** — the deliverable |

---

## OPEN — RIGHTEOUS STACK (Apex Spliceosome Organism)

> **FORM position:** Orthodox + Righteous NP-Complete. 34 tools, ~94 parameters, 10 layers. Built to be beaten by the deterministic stack. See [RIGHTEOUS_STACK.md](RIGHTEOUS_STACK.md).

| ID | Bounty | Status | FORM | TRUTH | Impact |
|----|--------|--------|------|-------|--------|
| RS1 | **VastDB + Ensembl atlas for 250 operator genes** — Isoform counts, NMD transcripts, retained introns, AS event types per operator. Database lookup, zero compute. | IN PROGRESS | Orthodox | Righteous | **CRITICAL** |
| RS2 | **ENCODE eCLIP lookup** — Which of 150 mapped RBPs bind preferentially to which operator's transcripts? Cross-reference eCLIP peaks with operator gene coordinates. | OPEN | Orthodox | Righteous | HIGH |
| RS3 | **SpliceVarDB query** — Known pathogenic splice variants in all 250 operator genes. Which operators are most vulnerable to splice-disrupting mutations? | OPEN | Orthodox | Righteous | HIGH |
| RS4 | **BRIE2 per-cell PSI** — Bayesian posterior with uncertainty (not point estimates). TIER 1 for 10x 3'. Focus on events <1kb from polyA. Flag imputed vs measured. | OPEN (needs BAM upload) | Orthodox | Righteous | **CRITICAL** |
| RS5 | **Leafcutter annotation-free differential** — Pseudobulk P vs S. Junction-only (no exon body). Dirichlet-multinomial. TIER 1 for our data. Blind to intron retention. | OPEN (needs BAM upload) | Orthodox | Righteous | **CRITICAL** |
| RS6 | **scShiba 3'-bias-aware analysis** — The only tool that explicitly corrects positional bias. Cross-validate with Leafcutter. TIER 1. New tool (2025), verify empirically. | OPEN (needs BAM upload) | Orthodox | Righteous | **CRITICAL** |
| RS7 | **MARVEL modality classification** — TIER 2. Point estimates are noisy from 10x 3'. Use for modality (bimodal/unimodal) not PSI values. Dip test unreliable at low coverage. | OPEN (needs BAM upload) | Orthodox | Righteous | HIGH |
| RS8 | **SpliZ deviation screen** — TIER 2. Quick candidate gene list. Tells you WHICH genes changed but not HOW. Follow up hits with BRIE2/Leafcutter. | OPEN (needs BAM upload) | Orthodox | Righteous | MEDIUM |
| RS9 | **rMATS-turbo pseudobulk JC-only** — TIER 2. JC mode ONLY (JCEC is catastrophically wrong with 3' bias). Create proper pseudobulk replicates (3 random splits per condition). | OPEN (needs BAM upload) | Orthodox | Righteous | HIGH |
| RS10 | **SpliceAI scores for operator exon boundaries** — Orthogonal: predicts from DNA sequence, not RNA-seq. Do RIBO genes have stronger splice sites than GOLGI? | OPEN | Rogue | Righteous | HIGH |
| RS11 | **IntEREst U12 annotation only** — Use the U12 intron gene list (695 introns). Do NOT trust IR ratios from 3' data (position-dependent bias). Cross-reference U12 genes with operators. | OPEN | Orthodox | Righteous | HIGH |
| RS12 | **CirComPara2 circRNA from BAMs** — Meta-pipeline (4 tools, consensus >= 2). Per-operator circRNA. Validates RIBO=0, GOLGI=2. May have low sensitivity from 3' data (circRNA lack polyA). | OPEN (needs BAM upload) | Orthodox | Righteous | MEDIUM |
| RS_AVOID | **SCSES diffusion imputation** — EVALUATED AND REJECTED. Fills sparse 3' PSI matrix with 90% imputation. Creates false confidence. Pretty matrix, mostly hallucinated. | REJECTED | — | — | — |
| RS10 | **Cross-reference: deterministic vs righteous** — For each metric, compare Daemon answer to Righteous answer. Convergence table. The TRUTH test. | OPEN (needs RS1-RS9) | Both | Both | **CRITICAL** |

---

## CONSTELLATION TARGET

The MarathonLament Spliceosome Catalogue becomes **Constellation: Spliceosome Atlas** when:
- ML9 (per-operator splice profile table) is complete
- ML12 (MaxEntScan mechanistic mapping) is complete
- ML20 (blackberry phone catalogue) is complete
- At least one disease splice atlas entry (ML13 or ML16) is complete
- Website architecture designed (SA5 from DiscordIntoSymphony bounty board)

---

*Front MarathonLament: the lament is not that the transcripts are broken. The lament is that we've been counting them as if they're all the same.*