Inheritance Detail Enrichment Project

Motivation

Mondo is actively debating how to represent the inherited vs. de novo distinction in the ontology (see #8074 on tuberous sclerosis, #8483 on genetic epilepsies). The dismech schema already has slots for de_novo_rate, penetrance, expressivity, and parent_of_origin_effect on Inheritance objects, but these are almost entirely unpopulated. Filling them in would:

1. Provide Mondo with structured, evidence-backed data on how often diseases arise de novo vs. are inherited, directly informing the #8483 proposal.
2. Surface cases where the inherited/de novo split is clinically meaningful (different severity, different recurrence risk) vs. cases where it is not.
3. Identify diseases where somatic mosaicism complicates the simple germline-inherited vs. germline-de-novo dichotomy.

Current State

The 4 files with de_novo_rate today: Achondroplasia (>90%), Achondrogenesis Type II (most cases), Temple-Baraitser (>90%), CINCA (>80%).

Priority Tiers

Tier 0: Fix the 10 Mendelian files missing inheritance entirely

These need at least mode-of-inheritance before detail fields make sense.

Tier 1: AD disorders likely to have high de novo rates

Autosomal dominant disorders with severe phenotypes (reduced reproductive fitness) are the most informative cases for the Mondo discussion - they are where the inherited/de novo distinction matters most for genetic counseling and recurrence risk.

Priority targets (AD, severe, high expected de novo fraction):

Tier 2: AD disorders with lower or unknown de novo fraction

Tier 3: AR / X-linked disorders

For autosomal recessive disorders, the de novo distinction is less clinically impactful (both alleles must be affected), but penetrance and expressivity data is still valuable. X-linked disorders have intermediate relevance (de novo rate in carrier mothers matters for counseling).

Data Sources

For each disorder, the curation agent should consult in order:

1. GeneReviews (via aurelian fulltext): Most GeneReviews entries have an "Inheritance" or "Molecular Genetics" section with de novo rates and penetrance. This is the single best source.

2. OMIM clinical synopsis: Often states "sporadic" or "de novo" with approximate rates. The phenotypic series pages are particularly useful for distinguishing inherited from de novo subtypes.

3. ClinGen gene-disease validity: Structured curation that includes inheritance evidence. Available at https://search.clinicalgenome.org/.

4. Primary literature (PubMed): For specific quantitative estimates. Large cohort studies or systematic reviews are preferred.

Fields to Populate

For each Inheritance entry:

All claims require evidence with exact PubMed abstract quotes per dismech SOP.

Special Cases for the Mondo Discussion

Some disorders don't fit a clean inherited/de novo binary. These are particularly informative for the #8483 proposal:

Somatic mosaicism complicates the picture

• Proteus syndrome (MONDO:0008318): Caused by somatic AKT1 mutations; never inherited (lethal if germline). This is neither "inherited" nor "de novo germline" - it's somatic-only.
• Tuberous sclerosis (#8074): Can be germline inherited, germline de novo, or somatic mosaic. The mosaic cases may have milder phenotypes and are not heritable.

De novo rate varies by gene within a phenotypic series

• Osteogenesis Imperfecta: OI type I (COL1A1/COL1A2) has lower de novo rates than OI type II (often de novo, lethal). Mondo groups these under OMIM phenotypic series - how should the de novo rate attach?
• Noonan Syndrome: PTPN11 mutations more often inherited; RAF1 mutations more often de novo. The de novo rate is gene-specific, not disease-level.

"De novo" means different things

• Germline de novo in the parent's gamete (classic)
• Post-zygotic de novo leading to mosaicism
• Somatic mutation in affected tissue only

The de_novo_rate field as currently defined conflates these. We may want to add a de_novo_type enum or similar.

Curation Protocol

Per-disorder workflow

1. Identify the GeneReviews entry (if it exists) using aurelian fulltext.
2. Extract inheritance details: mode, penetrance, de novo rate, expressivity, parent-of-origin effects.
3. For each claim, find a PubMed-cited sentence that can serve as an exact-quote snippet.
4. If GeneReviews lacks specifics, search OMIM and primary literature.
5. Populate the Inheritance block with structured fields + evidence.
6. Run just validate kb/disorders/<file>.yaml and just validate-kb-references kb/disorders/<file>.yaml.

Batch approach

An AI curation agent can draft inheritance blocks for multiple disorders in a batch, but each must be validated against the reference validation pipeline before merge. Estimated effort:

• Tier 0 (10 files, basic mode): ~1 agent session
• Tier 1 (14 disorders, full detail): ~2-3 agent sessions
• Tier 2 (9 disorders): ~1-2 agent sessions
• Tier 3 (remaining AR/XL): ~2-3 agent sessions

Output for Mondo

Once populated, generate a summary table for the Mondo #8483 discussion:

MONDO_ID | Disease | Mode | De_Novo_Rate | Penetrance | Mosaicism? | Source

This table would concretely demonstrate: - Which diseases have a clinically meaningful inherited/de novo split - Where the binary fails (mosaicism, gene-specific rates) - What structured fields an ontology would need to capture this data

Schema Considerations

The current schema handles most cases, but two extensions may be needed:

1. de_novo_type enum: To distinguish germline de novo, post-zygotic mosaic, and somatic-only. Values might be: GERMLINE_DE_NOVO, POSTZYGOTIC_MOSAIC, SOMATIC_ONLY, MIXED, UNKNOWN.

2. Gene-specific inheritance blocks: Some disorders need per-gene inheritance data (e.g., Noonan: PTPN11 vs. RAF1 have different de novo rates). The current schema attaches inheritance to the disease, not to the gene. A gene slot on Inheritance (or nesting inheritance under genetic) might be needed.