Lifesciences Reporting

Life Sciences Reporting

Format professional reports from Fuzzy-to-Fact pipeline output using domain-specific templates.

Grounding Rule

ALL claims in the report MUST trace to specific tool calls from Phases 1-5.
Do NOT introduce new entities, drug names, gene functions, or trial IDs from
training knowledge during report generation. The report SYNTHESIZES existing
Phase 1-5 output — it does NOT generate new facts.

If a Phase returned no results for a section, report "No data retrieved" with
the tool name that was called. Do NOT fill gaps from memory.

Scope

In scope: Template selection, evidence grading, source attribution, formatting, synthesis of Phase 1-5 output.

Out of scope: API calls, data retrieval, new entity resolution, creating new Fuzzy-to-Fact phases, modifying other skills. This skill consumes pipeline output — it does not produce it.

Template Decision Tree

Route the user's query to the appropriate template using this priority order. If multiple categories apply, combine sections from relevant templates (see Multi-Template Combination below).

1. HOW does a drug work?              --> Template 6: Mechanism Elucidation
2. Drug SAFETY or off-targets?        --> Template 7: Safety / Off-Target
3. REGULATORY milestones or filings?  --> Template 5: Regulatory / Commercialization
4. FIND or REPURPOSE drugs?           --> Template 1: Drug Discovery / Repurposing
5. GENE/PROTEIN interactions?         --> Template 2: Gene / Protein Network
6. CLINICAL TRIALS broadly?           --> Template 3: Clinical Landscape
7. VALIDATE a target?                 --> Template 4: Target Validation
8. Multiple categories?               --> Combine sections from relevant templates

Competency Question Coverage

Template 1: Drug Discovery / Repurposing

Use for: cq2 (FOP repurposing), cq4 (AD therapeutics), cq7 (NGLY1 multi-hop), cq8 (ARID1A SL), cq10 (HD novel targets), cq14 (TP53 SL).

## Summary
[Direct answer: what drugs were found and why they are relevant]

## Resolved Entities
| Entity | CURIE | Type | Source |
|--------|-------|------|--------|
| [name] | [CURIE] | Gene/Protein/Disease | [Source: tool(param)] |

## Drug Candidates
| Drug | CURIE | Phase | Mechanism | Target | Evidence Level | Source |
|------|-------|-------|-----------|--------|---------------|--------|
| [name] | [CURIE] | [1-4] | [action] | [gene] | [L1-L4] | [Source: tool(param)] |

## Mechanism Rationale
[For each drug: why this drug targets the disease pathway.
Trace: Drug --[mechanism]--> Target --[pathway]--> Disease]

## Clinical Trials
| NCT ID | Title | Phase | Status | Verified | Source |
|--------|-------|-------|--------|----------|--------|
| [ID] | [title] | [phase] | [status] | [Y/N] | [Source: tool(param)] |

## Evidence Assessment
[Claim-level grades using the Evidence Grading System below]

## Gaps and Limitations
[What was NOT found; which tools returned no results]

Template 2: Gene / Protein Network

Use for: cq3 (AD gene networks), cq5 (MAPK cascade), cq6 (BRCA1 regulatory network).

## Summary
[Direct answer: what network was found and its biological significance]

## Resolved Entities
| Entity | CURIE | Type | Source |
|--------|-------|------|--------|
| [name] | [CURIE] | Gene/Protein | [Source: tool(param)] |

## Interaction Network
| Protein A | Protein B | Score | Type | Direction | Source |
|-----------|-----------|-------|------|-----------|--------|
| [name] | [name] | [0-1000] | [physical/regulatory] | [A->B / bidirectional] | [Source: tool(param)] |

## Hub Genes
| Gene | Degree | Key Interactions | Disease Associations | Source |
|------|--------|------------------|---------------------|--------|
| [name] | [N] | [top partners] | [diseases] | [Source: tool(param)] |

## Pathway Membership
| Pathway | ID | Member Genes from Query | Source |
|---------|-----|------------------------|--------|
| [name] | [WP:ID] | [gene list] | [Source: tool(param)] |

## Network Properties
- Total nodes: [N]
- Total edges: [N]
- Average interaction score: [N]
- Regulatory vs physical: [ratio]

## Evidence Assessment
[Claim-level grades]

## Gaps and Limitations
[Missing interactions, low-confidence edges, unresolved entities]

Template 2 Notes

• Disease CURIE: Not required in Resolved Entities table unless drug/trial discovery was performed
• Pathway Membership: REQUIRED section (use WikiPathways for all core genes)
• Clinical Trials: Only include if relevant to the comparative network question
• Source Attribution: Paraphrasing UniProt function text is acceptable; cite the tool call

Template 3: Clinical Landscape

Use for: cq12 (health emergencies 2026).

## Summary
[Direct answer: what clinical trial patterns were found]

## Phase Distribution
| Phase | Count | Top Conditions | Source |
|-------|-------|---------------|--------|
| Phase 3 | [N] | [conditions] | [Source: tool(param)] |
| Phase 2 | [N] | [conditions] | [Source: tool(param)] |
| Phase 1 | [N] | [conditions] | [Source: tool(param)] |

## Recruiting Trials
| NCT ID | Condition | Intervention | Phase | Sponsor | Source |
|--------|-----------|-------------|-------|---------|--------|
| [ID] | [condition] | [drug/device] | [phase] | [org] | [Source: tool(param)] |

## Therapeutic Trends
| Trend | Trial Count | Representative Trials | Source |
|-------|------------|----------------------|--------|
| [e.g., CAR-T expansion] | [N] | [NCT IDs] | [Source: tool(param)] |

## Evidence Assessment
[Claim-level grades]

## Gaps and Limitations
[Search coverage, date range caveats, missing trial types]

Template 4: Target Validation

Use for: cq11 (p53-MDM2-Nutlin axis).

## Summary
[Direct answer: is this target validated and druggable?]

## Target Profile
| Property | Value | Source |
|----------|-------|--------|
| Gene Symbol | [symbol] | [Source: tool(param)] |
| CURIE | [HGNC:ID] | [Source: tool(param)] |
| Protein | [UniProt ID] | [Source: tool(param)] |
| Function | [text from UniProt] | [Source: tool(param)] |
| Biotype | [protein_coding/etc] | [Source: tool(param)] |

## Disease Associations
| Disease | Score | Evidence Sources | Source |
|---------|-------|-----------------|--------|
| [name] | [0-1] | [genetic, literature, etc] | [Source: tool(param)] |

## Tractability Assessment
| Modality | Label | Value | Source |
|----------|-------|-------|--------|
| Small molecule | [Clinical Precedence/etc] | [score] | [Source: tool(param)] |
| Antibody | [label] | [score] | [Source: tool(param)] |

## Known Drugs
| Drug | Phase | Mechanism | Source |
|------|-------|-----------|--------|
| [name] | [1-4] | [action] | [Source: tool(param)] |

## Interaction Partners
| Partner | Score | Type | Source |
|---------|-------|------|--------|
| [name] | [0-1000] | [physical/regulatory] | [Source: tool(param)] |

## Evidence Assessment
[Claim-level grades]

## Gaps and Limitations
[Missing tractability data, low-confidence associations]

Template 5: Regulatory / Commercialization

Use for: cq13 (high-commercialization trials), cq15 (CAR-T regulatory).

## Summary
[Direct answer: which trials or drugs have highest commercial/regulatory momentum?]

## Resolved Entities
| Entity | CURIE | Type | Source |
|--------|-------|------|--------|
| [name] | [CURIE] | Drug/Trial | [Source: tool(param)] |

## Milestone Timeline
| Trial/Drug | Event | Date/Status | Significance | Source |
|-----------|-------|-------------|-------------|--------|
| [name] | [Phase 3 initiation/BLA filing/etc] | [date] | [first-in-class/etc] | [Source: tool(param)] |

## Competitive Landscape
| Target | Drug | Sponsor | Phase | Status | Source |
|--------|------|---------|-------|--------|--------|
| [target] | [drug] | [company] | [phase] | [recruiting/completed] | [Source: tool(param)] |

## Investment Signals
| Signal | Trial/Drug | Evidence | Source |
|--------|-----------|----------|--------|
| [large enrollment/breakthrough designation/etc] | [name] | [detail] | [Source: tool(param)] |

## Evidence Assessment
[Claim-level grades]

## Gaps and Limitations
[Regulatory data not in ClinicalTrials.gov, proprietary deal data unavailable]

Template 6: Mechanism Elucidation

Use for: cq1 (Palovarotene mechanism for FOP).

## Summary
[Direct answer: by what mechanism does Drug X treat Disease Y?]

## Mechanism Chain
[Narrative tracing the path: Drug --[action]--> Target --[regulation]--> Pathway --[association]--> Disease]

### Step-by-Step
| Step | From | Relationship | To | Evidence | Source |
|------|------|-------------|-----|----------|--------|
| 1 | [Drug] | [agonist/inhibitor] | [Target protein] | [mechanism data] | [Source: tool(param)] |
| 2 | [Target] | [regulates/inhibits] | [Downstream] | [interaction data] | [Source: tool(param)] |
| 3 | [Downstream] | [associated_with] | [Disease] | [association data] | [Source: tool(param)] |

## Supporting Evidence
| Claim | Evidence Level | Sources |
|-------|---------------|---------|
| [Drug is agonist of Target] | [L3] | [Source: tool(param)] |
| [Target regulates Downstream] | [L2] | [Source: tool(param)] |

## Alternative Mechanisms
[Other proposed mechanisms from literature, if retrieved]

## Evidence Assessment
[Claim-level grades]

## Gaps and Limitations
[Missing pathway steps, unconfirmed regulatory direction]

Template 7: Safety / Off-Target

Use for: cq9 (Dasatinib off-target risks).

## Summary
[Direct answer: what are the safety risks of Drug X?]

## Index Compound Profile
| Property | Value | Source |
|----------|-------|--------|
| Drug Name | [name] | [Source: tool(param)] |
| CURIE | [CHEMBL:ID] | [Source: tool(param)] |
| Primary Target(s) | [target list] | [Source: tool(param)] |
| Approved Indication(s) | [diseases] | [Source: tool(param)] |

## Off-Target Hits
| Off-Target | Gene CURIE | Activity (IC50/Ki) | Clinical Consequence | Source |
|-----------|-----------|-------------------|---------------------|--------|
| [protein] | [HGNC:ID] | [nM] | [e.g., cardiotoxicity] | [Source: tool(param)] |

## Selectivity Comparison
| Target | Drug X (IC50) | Comparator Drug (IC50) | Selectivity Ratio | Source |
|--------|--------------|----------------------|-------------------|--------|
| [primary] | [nM] | [nM] | [ratio] | [Source: tool(param)] |
| [off-target] | [nM] | [nM] | [ratio] | [Source: tool(param)] |

## Safety Signals
| Signal | Mechanism | Severity | Frequency | Source |
|--------|-----------|----------|-----------|--------|
| [e.g., QT prolongation] | [hERG inhibition] | [serious] | [common/rare] | [Source: tool(param)] |

## Evidence Assessment
[Claim-level grades]

## Gaps and Limitations
[Missing activity data, untested off-targets, post-market data not available]

Evidence Grading System

Grade each claim individually, then compute an overall report confidence. This prevents inflated "high confidence" when some claims are grounded and others are not.

Evidence Levels

Modifiers

Grading Procedure

For each claim in the report:

1. Identify the base level (L1-L4) from the criteria above
2. Apply all applicable modifiers (sum adjustments, clamp to 0.00-1.00)
3. Record the final score and justification

For the overall report:

• Compute the median of all claim scores (not the mean — resistant to outliers)
• Report the range (lowest to highest claim score)
• Flag any claim below L1 (0.30) as "Insufficient Evidence"

Worked Example: Venetoclax/BCL2

Claim: "Venetoclax inhibits BCL2 and is approved for CLL"

1. Base: L4 (0.90) — FDA-approved drug for this indication
2. Modifiers:
   • Active trials: +0.10 (multiple recruiting CLL trials found via clinicaltrials_search_trials)
   • Multi-DB: already captured in L4 base
3. Final: 0.95 (L4 Clinical)
4. Sources: [Source: opentargets_get_target(ENSG00000171791)], [Source: clinicaltrials_search_trials("venetoclax CLL")]

Claim: "BCL2 interacts with BAX (STRING score 0.999)"

1. Base: L2 (0.55) — confirmed by STRING
2. Modifiers:
   • High STRING score: +0.05 (score >= 900)
   • UniProt function text mentions BAX: +0.05 (literature support)
3. Final: 0.65 (L2 Multi-DB)
4. Sources: [Source: string_get_interactions(9606.ENSP00000381185)], [Source: uniprot_get_protein(Q07817)]

Source Attribution Standards

Every factual claim must cite the tool call that produced it. Formats:

Formatting Standards

Table Conventions

• CURIE column: always present for entities, using full CURIE format (HGNC:11998, not 11998)
• Source column: always the rightmost column
• Evidence Level column: use shorthand L1-L4 in tables, expand in Evidence Assessment section
• Sort drug candidate tables by Phase (descending), then Evidence Level (descending)
• Sort interaction tables by Score (descending)

First-Mention Rule

On first mention of any entity, include both the human-readable name and CURIE:

TP53 (HGNC:11998) encodes the p53 tumor suppressor protein (UniProtKB:P04637).

Subsequent mentions may use the name alone.

Markdown & General

• ## H2 for major sections, ### H3 for subsections, #### H4 sparingly
• Tables for structured data; prose for narrative synthesis
• No emoji; dates in ISO 8601; scores to 2 decimal places; IC50/Ki in nM
• Abbreviations: define on first use, then abbreviate

Common Report Pitfalls

Hallucination Injection

Problem: Reporting fills in "expected" drug names or trial IDs from training knowledge when Phase 4a or 4b returned sparse results.

Prevention: Every row in Drug Candidates and Clinical Trials tables must have a [Source: tool(param)] citation. If no source exists, the row must not appear. Write "No drug candidates retrieved" rather than guessing.

Inflated Confidence

Problem: Assigning L3/L4 evidence to claims supported by a single STRING interaction.

Prevention: Apply grading procedure strictly. A single-database claim is L1 (0.30-0.49) regardless of how "well known" the relationship seems. Modifiers can raise it, but only based on actual tool output.

Wrong Template Selection

Problem: Using Drug Discovery template for a query about gene networks (cq3, cq5, cq6), producing empty Drug Candidates and Clinical Trials sections.

Prevention: Follow the decision tree. If the query mentions "interact", "regulate", "network", or "cascade" without mentioning drugs, use Template 2 (Gene/Protein Network).

Unverified NCT IDs

Problem: Including NCT IDs from Phase 4b search results without Phase 5 verification.

Prevention: The "Verified" column in Clinical Trials tables must reflect Phase 5 output. If Phase 5 was not run, mark as "Unverified" (not "Yes").

Mechanism Mismatch Blindness

Problem: For gain-of-function diseases, including agonists in Drug Candidates without flagging the mechanism conflict.

Prevention: Cross-reference the disease biology (from Phase 2 enrichment) with each drug's mechanism. Flag agonists for gain-of-function diseases with a -0.20 modifier and a note in Gaps and Limitations.

Paraphrasing vs Hallucination Confusion

Problem: Reviewers flag faithful paraphrasing of UniProt function text as "hallucination" because the report text doesn't match verbatim.

Acceptable synthesis (not hallucination):

• UniProt function text paraphrased for readability: "Binds to 3 E-boxes of the E-cadherin/CDH1 gene promoter" → "binds E-boxes in CDH1 promoter"
• Multiple tool outputs synthesized into coherent narrative with all sources cited
• Interpretive claims clearly marked with qualifiers like "[Inferred from...]"

Unacceptable (hallucination):

• Entity names, CURIEs, or NCT IDs not present in tool outputs
• FDA approval years (e.g., "FDA-approved 2021"), prevalence statistics (e.g., ">50%"), or trial outcome numbers without sources
• Mechanistic details that extend beyond what tool output states

Best practice: Add a synthesis disclaimer to reports that paraphrase extensively:

"Mechanism descriptions paraphrase UniProt function text and STRING interaction annotations. All synthesis is grounded in cited tool calls; no entities, CURIEs, or quantitative values are introduced from training knowledge."

Combining Templates Poorly

Problem: Including all sections from two templates, creating a bloated report with redundant entity tables.

Prevention: Share Resolved Entities and Evidence Assessment across templates. Include unique sections from each. See Multi-Template Combination below.

Multi-Template Combination

When a query spans multiple templates (e.g., cq2 involves both Drug Discovery and Mechanism Elucidation), combine them:

Shared Sections (include once)

• Summary
• Resolved Entities
• Evidence Assessment
• Gaps and Limitations

Template-Specific Sections (include from each relevant template)

Take the distinctive sections from each template. Do not duplicate entity tables.

Worked Example: cq2 (FOP Drug Repurposing)

Query: "What drugs targeting the BMP pathway could be repurposed for FOP?"

Templates: Drug Discovery (primary) + Mechanism Elucidation (secondary)

Report structure:

1. Summary (shared)
2. Resolved Entities (shared)
3. Drug Candidates (from Template 1)
4. Mechanism Chain (from Template 6 — traces BMP pathway logic)
5. Mechanism Rationale (from Template 1 — per-drug justification)
6. Clinical Trials (from Template 1)
7. Evidence Assessment (shared — covers all claims)
8. Gaps and Limitations (shared)

See Also

• lifesciences-graph-builder: Orchestrator for full Fuzzy-to-Fact protocol (produces the data this skill formats)
• lifesciences-genomics: HGNC, Ensembl, NCBI gene resolution (Phase 1-2 data sources)
• lifesciences-proteomics: UniProt, STRING, BioGRID interaction data (Phase 2-3 data sources)
• lifesciences-pharmacology: ChEMBL, PubChem, IUPHAR, Open Targets drug data (Phase 4a data sources)
• lifesciences-clinical: Open Targets associations, ClinicalTrials.gov (Phase 4b-5 data sources)
• lifesciences-crispr: BioGRID ORCS essentiality validation (Phase 3 extension)