transhumanist-already-exists/human-gene-lof-rescue-evidence

Human Biallelic Loss-of-Function and Functional-Rescue Evidence

This dataset contains 88 curated human gene records linking three experimentally distinct observations:

1. 1.biallelic human loss of function;
2. 2.a consistent phenotype reported in independent affected families or cohorts;
3. 3.functional rescue in affected humans or patient-derived human cells.

Each record therefore connects genotype → recurrent human phenotype → reversal of a disease-relevant defect. This convergent evidence is unusually useful for causal biological reasoning because the direction of the perturbation is known and the altered phenotype can be tested for reversibility.

The dataset is intended for research on mechanistic biological reasoning, gene-function interpretation, evidence retrieval, and model evaluation. It is not a clinical diagnostic resource.

Why these records are biologically valuable

Most gene–phenotype datasets mix evidence of very different strength: differential expression, pathway enrichment, animal phenotypes, single-family observations, and direct perturbation experiments. Those sources answer different questions. The records here were retained only when the evidence chain contained all of the following components.

1. The perturbation occurred in humans

Affected people carried biallelic variants consistent with loss of gene function. The causal direction is therefore interpretable: reduced gene function precedes the disease phenotype. Gain-of-function and dominant-negative mechanisms were outside the scope of this release.

2. The phenotype recurred independently

The core phenotype had to recur in affected people from at least two primary human-genetics publications representing distinct families and substantially different author groups. Recurrence across independent genetic backgrounds reduces the chance that the observation reflects a private family background, an unrelated segregating variant, or a single ascertainment error.

3. Restoring function reversed a relevant defect

The evidence had to include functional complementation or rescue. Eligible experiments included introduction of a wild-type gene into patient-derived cells, correction of the causal allele, replacement of the missing gene product, natural somatic reversion, or human gene therapy. The measured endpoint had to represent a disease-relevant cellular, biochemical, physiological, or clinical abnormality. Restoration of expression alone was insufficient.

Rescue adds specificity to the genetic observation. If loss of a gene and the phenotype merely co-occur, another linked variant or background effect can remain plausible. If restoration of the same gene or gene product reverses the corresponding defect, the proposed gene-to-phenotype mechanism receives direct experimental support.

4. The evidence converges across biological scales

The retained records combine human Mendelian genetics with a mechanistic assay. Depending on the disease, the rescued endpoint may be enzyme activity, metabolite accumulation, protein trafficking, organelle function, DNA repair, immune-cell signaling, cellular differentiation, tissue-level function, or a clinical phenotype after therapy. This makes the records useful as grounded examples of how a molecular lesion propagates toward a cellular and human phenotype.

These criteria provide strong causal support within the stated disease and experimental context. They do not establish that every clinical manifestation is caused through the assayed cellular endpoint, or that a cellular rescue would reverse established disease in a person.

Dataset contents

One datapoint

{
  "datapoint_id": "human-lof-rescue-abhd5",
  "language": "en",
  "gene": "ABHD5",
  "hgnc_id": "HGNC:21396",
  "diseases": ["Dorfman-Chanarin disease"],
  "evidence_profile": {
    "human_biallelic_loss_of_function": true,
    "independent_human_genetic_reports": true,
    "consistent_core_phenotype": true,
    "functional_rescue_in_humans_or_patient_derived_cells": true
  },
  "claim_provenance": {
    "origin": "model-generated normalization reviewed against ClinGen evidence summaries and linked article text",
    "verbatim": false,
    "translation": false,
    "warning": "Normalized claims must not be presented as quotations from the cited articles."
  },
  "normalized_claims": {
    "loss_of_function": "Autosomal-recessive ABHD5 loss of function impairs cellular lipolysis.",
    "human_phenotype_consistency": "Independent reports describe congenital ichthyosis and multisystem triglyceride accumulation.",
    "functional_rescue": "Wild-type ABHD5 expression in patient fibroblasts restored lipolysis and reduced abnormal triacylglycerol accumulation.",
    "causal_summary": "ABHD5 loss of function causes defective triglyceride mobilization; wild-type ABHD5 reverses the corresponding cellular phenotype."
  },
  "evidence_chunks": {
    "human_genetics": [
      {
        "pmid": "11590543",
        "chunk_id": "chunk-abhd5-human-genetics-pmid-11590543",
        "chunk_link": "article_sources/SOURCE_CHUNKS.md#chunk-abhd5-human-genetics-pmid-11590543",
        "article_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC1274347/",
        "chunk_type": "pubmed_abstract_sentence",
        "section": "Abstract",
        "position": 1
      }
    ],
    "functional_rescue": [
      {
        "pmid": "16679289",
        "chunk_id": "chunk-abhd5-functional-rescue-pmid-16679289",
        "chunk_link": "article_sources/SOURCE_CHUNKS.md#chunk-abhd5-functional-rescue-pmid-16679289",
        "article_url": "https://pubmed.ncbi.nlm.nih.gov/16679289/",
        "chunk_type": "pubmed_abstract_sentence",
        "section": "Abstract",
        "position": 8
      }
    ]
  }
}

The normalized_claims fields are concise English syntheses. They are not translations and must not be treated as quotations from the papers. Exact short locator excerpts are stored separately in data/source_chunks.jsonl and article_sources/SOURCE_CHUNKS.md.

Evidence standard

A retained record had to satisfy all of the following project-specific criteria:

• —a human autosomal-recessive disease mechanism caused by biallelic loss of function;
• —a consistent core phenotype in at least two primary human-genetics publications representing independent affected families and substantially distinct author groups;
• —restoration using a wild-type gene, correction of the causal variant, replacement of the missing gene product, natural somatic reversion, or successful gene therapy in affected humans or patient-derived human cells;
• —a disease-relevant rescue endpoint rather than expression alone or an unrelated reporter;
• —no PubMed retraction or correction signal undermining the evidence chain.

Functionally demonstrated hypomorphic loss-of-function variants were eligible. Gain-of-function and dominant-negative mechanisms were excluded.

Screening flow

The final review excluded three additional records that passed the structured adjudications: a non-human rescue model for MPDU1, an insufficiently disease-relevant rescue endpoint for RELB, and an indirect endpoint for SLC39A13.

Source-chunk coverage

The 88 datapoints contain 276 evidence references covering 254 unique PubMed records:

• —67 references point to locally indexed PMC full-text paragraphs;
• —207 references point to PubMed abstract sentences;
• —2 references are metadata-only because no usable article text was available in the ingestion sources.

The metadata-only references are the FANCF rescue paper (PMID 10615118) and one NAXD human-genetics paper (PMID 32462209). These records remain visibly flagged and must not be interpreted as text-grounded chunks.

Chunk selection was lexical and should be manually verified before using the excerpt itself as a gold evidence span. The relationship-level inclusion decision was based on the broader ClinGen evidence dossier, PubMed metadata, two structured adjudications, and final dossier review.

Loading

from datasets import load_dataset

dataset = load_dataset(
    "transhumanist-already-exists/human-gene-lof-rescue-evidence",
    split="train",
)
print(dataset[0])

Important limitations

• —This is a high-precision subset of the ClinGen source frame rather than an exhaustive catalog of qualifying human genes.
• —ClinGen evidence summaries and PubMed metadata were the principal screening substrate. The release does not replace a complete manual extraction of every primary paper's methods and results tables.
• —Author-set overlap and distinct affected families were used to assess replication. Hidden cohort or consortium dependence can remain.
• —A cellular rescue supports a causal gene-to-cellular-phenotype relationship. It does not establish reversibility of every organ-level clinical manifestation.
• —Normalized claims were generated with language-model assistance and reviewed against the structured evidence dossiers. Users should follow the linked evidence before treating a claim as a benchmark label.
• —Absence from the dataset does not imply that a gene–disease relationship is false.

Licensing and attribution

The dataset's normalized annotations, curation tables, and code may be reused with attribution to this dataset. Original publications, abstracts, and source excerpts remain subject to their publishers' and authors' respective terms. The short excerpts are provided as evidence locators; consult the linked source for reuse rights and full context.

Suggested citation

@dataset{human_gene_lof_rescue_evidence_2026,
  title  = {Human Biallelic Loss-of-Function and Functional-Rescue Evidence},
  author = {transhumanist-already-exists},
  year   = {2026},
  url    = {https://huggingface.co/datasets/transhumanist-already-exists/human-gene-lof-rescue-evidence}
}