This page documents the meaning of each fileds (columns) used in the variant databases available in PubMind.
For the column that has | in it, it means the aggregated value for each record of a variant. The order of different columns should match. For example, the order in column formatted_reference should match with LLM_reasoning and pathogenicity.
PVID: Unique PubMind Variant ID.gene: Name of the gene extracted by LLM.LLM_mutation: Raw mutation text from the LLM, before parsing or normalization. Kept for transparency and as a fallback when those steps fail.formatted_reference: Citation(s) or source(s) of the variant.MONDO_name_09: Normalized disease name assigned based on 90% similarity with MONDO human disease database.MONDO_ID_09: Corresponding MONDO ID for MONDO_name_09.LLM_reasoning: Reasoning provided by the LLM for pathogenicity classification.pathogenicity: LLM-assigned classification (i.e., pathogenic, likely pathogenic, benign, likely benign, unknown).disease: Disease name extracted from literature paragraph by LLM before normalization (for SNV).related_disease: Disease name extracted from literature paragraph by LLM before normalization (for complex variants).Num_of_record_used: Number of individual records (paragraphs) supporting/mentioning this variant.Num_of_paper_used: Number of individual papers supporting/mentioning this variant.pathogenicity_sum: Final pathogenicity assignment across all records.pathogenicity_score: Quantitative score summarizing the pathogenicity across all records (range: 0–1).paper_level_pathogenicity_score: Quantitative score summarizing pathogenicity across papers at the paper level (range: 0–1).pathogenicity_score_per_paper: Per-paper pathogenicity scores used for paper-level aggregation.confidence: Confidence level for a variant entry based on how many evidences are collected (range: 0-3).confidence_criteria: Reasoning or rules behind the confidence level.RSID: Reference SNP ID from dbSNP.dna_change: cDNA-level mutation (e.g., 123A>T).aa_change: Protein-level mutation (e.g., Glu41Lys).genomic_coord_result: Genomic coordinate(s) inferred from Ensembl transcript mapping. One variant could correspond to multiple transctips.parsed_variants: Parsed structured representation of the variant based on genomic_coord_result.phenotype: Phenotype terms extracted from literature paragraph by LLM before normalization.HPO_term_09: Normalized phenotype term assigned based on 90% similarity with HPO database.HPO_ID_09: Corresponding HPO phenotype ID for HPO_term_09.MONDO_name_counted: Aggregated number of MONDO disease names mentioned across records.HPO_term_counted: Aggregated number of HPO terms mentioned across records.MONDO_ID_counted: Aggregated number of MONDO disease IDs mentioned across records.HPO_ID_counted: Aggregated number of HPO IDs mentioned across records.PMCID_PMID_counted: Aggregated number of references (PMCID/PMID) for the variant.gene_fusion: Formatted fusion (e.g. EML4::ALK).first_gene: First gene of the fusion assigned by LLM, normally this is the driver gene in the fusion.partner_gene: Second gene of the fusion assigned by LLM, normally this is the partner gene.protein_domains_affected: Functional domains impacted, normally for first_gene.type: Structural variant type (deletion, duplication, inversion, etc.).chr_start_end: Genomic coordinates.gene_affected: Genes overlapped/affected by the SV.C112R, Cys112Arg, rs429358 — and each
spelling is recorded as its own variant with its own PVID, so all of them map to the
same coordinates. Papers may also assign one position to different overlapping genes.
Of 1,217,126 distinct alleles, 128,164 (10.5%) carry more than one PVID; the most is
57, at the APOE ε4 position.
Each row therefore reports the pathogenicity of that variant record, not of
the position as a whole, and two rows at the same coordinates can disagree.
hg38_pubmind_db_v111) holds one row per genomic position:
the PVIDs are listed together and pathogenicity_sum,
paper_level_pathogenicity_score and confidence are
recalculated across every underlying paper record, using the same rules as this site.
Download it from
ANNOVAR.
Chr: Chromosome on which the variant is located.Start: Start position of the variant on the chromosome (1-based coordinate).End: End position of the variant on the chromosome.Ref: Reference allele observed in the reference genome.Alt: Alternate allele identified in the variant record.MANE_transcript_used: If MANE Select transcript was used for the genomic coordinate. (True, False, RSID)Func.ensGene, Gene.ensGene, GeneDetail.ensGene,
ExonicFunc.ensGene, AAChange.ensGene,
CLNALLELEID (clinvar_20260829), CLNDN (clinvar_20260829), CLNDISDB (clinvar_20260829),
CLNREVSTAT (clinvar_20260829), CLNSIG (clinvar_20260829), gnomad41_genome_AF, and gnomad41_genome_AF_grpmax.
Please refer to the ANNOVAR documentation for detailed definitions of each annotation field.
Deleterious: Number of in-silico prediction tools classifying the variant as deleterious based on dbnsfp47a.Tolerated: Number of in-silico prediction tools classifying the variant as tolerated based on dbnsfp47a.Unknown: Number of in-silico prediction tools returning unknown or missing results based on dbnsfp47a.avg_rankscore: Average pathogenicity rank score aggregated from multiple prediction tools (scaled 0–1) based on dbnsfp47a.PubMind-DB provides summary-level API endpoints for programmatic access. The API returns aggregate counts, pathogenicity scores, and links back to the PubMind-DB website. Detailed per-record information is not returned through the API; please use the PubMind-DB website for detailed records.
Database release. Every response carries a db_version field naming the release that
produced it. Without a version parameter the API serves the current release
(1.1.1), which means a request will move to the next release when one ships. Add
version= to pin a release — recommended for anything reproducible, such as results cited in a
manuscript. Available releases: 1.1.1, 1.0.0.
An unrecognised value returns 400 with the list of valid releases rather than quietly falling back
to the default.
curl "https://pubmind.wglab.org/api/summary?query=ACVR1&field=gene&version=1.1.1"
curl "https://pubmind.wglab.org/api/pvid/PVID118384?version=1.1.1"
A v1.0.0 PVID that no longer exists in the current release does not simply fail: the API answers
301 with new_PVID and a retry_url if it was renumbered, or
410 with status: retired if the variant is no longer reported.
/api/summary: Summary search by gene, MONDO ID, MONDO disease name, PubMed ID, or PMC ID.
Supported field values include gene, MONDO_ID_09,
MONDO_name_09, and formatted_reference. Aliases such as
pmid, pmcid, mondo_id, and mondo_name are also accepted.
/api/pvid/<PVID>: Variant summary lookup by PubMind Variant ID.
Example: summary search for the BRCA1 gene.
curl "https://pubmind.wglab.org/api/summary?query=BRCA1&field=gene"The same request can be opened directly in a browser:
https://pubmind.wglab.org/api/summary?query=BRCA1&field=geneExample: summary lookup for PVID118384.
curl "https://pubmind.wglab.org/api/pvid/PVID118384"The same PVID request can be opened directly in a browser:
https://pubmind.wglab.org/api/pvid/PVID118384v100_PVID, variant_type, status (current, redirect, split or retired), v111_PVID, note. Old IDs still resolve on the site and the API, which redirect to the current PVID. Per-type counts are on the What's New page.If you have questions or suggestions, please visit the GitHub repo.