About Sugi Variant
A deterministic, cross-sourced reference for every germline human variant in ClinVar — open and free.
Sugi Variant assembles one reference view per germline human variant by deterministically mining the BioBTree knowledge graph. Anchored on ClinVar, each view brings together the clinical classification and condition, population frequency and gene constraint (gnomAD v4.1), calibrated predictors (AlphaMissense, REVEL, SaProt, evolutionary conservation, and SpliceAI for splice regions), and — where a ClinGen expert panel has curated the variant — the applied ACMG criteria, verbatim. Every fact is shown with its source and dataset build date, across all variant classes, including the non-coding and mitochondrial variants many tools omit.
Deterministic, not generated. Nothing on a view is written by a language model. Each is assembled the same way on every build by a fixed pipeline, so the reference is reproducible and every number is traceable to its source and version.
It describes evidence; it does not compute a verdict. Sugi Variant reports what the authoritative sources say and never re-tallies a classification of its own. Because computational predictors are correlated, one calibrated predictor carries the evidence weight (AlphaMissense for missense, conservation otherwise) and the rest are shown as agreement signals that concur or dissent, never as additive votes — following ClinGen's Sequence Variant Interpretation guidance.
The disagreement view. Laying the evidence out transparently lets Sugi Variant surface what a single score hides: the variants where the independent signals disagree — a predictor at odds with the clinical call, predictors split among themselves, or predictors unanimous where ClinVar's submitters conflict. These are a quality-control signal for review, not a reclassification.
Sugi Variant is the per-variant companion to the deterministic Sugi Atlas. Its pipeline and evaluation are described in the Method.
Reference and research use only. Sugi Variant is not a diagnostic device and is not medical advice. Interpretation of a variant for an individual must be done by a qualified clinician or genetic counsellor.
The site is free and open for everyone. For a custom build on your own data, get in touch.
Sources
Sugi Variant draws these datasets through BioBTree, which integrates them. Each fact is traced to its source.
| Source | Reference |
|---|---|
| ClinVar | Landrum et al., NAR 2018 |
| gnomAD v4.1 | Chen et al., Nature 2024 |
| AlphaMissense | Cheng et al., Science 2023 |
| REVEL | Ioannidis et al., AJHG 2016 |
| SaProt | Su et al., ICLR 2024 |
| SpliceAI | Jaganathan et al., Cell 2019 |
| CADD | Rentzsch et al., NAR 2019 |
| Conservation (phyloP/phastCons/GERP) | Pollard et al. 2010; Davydov et al. 2010 |
| gnomAD Non-coding Constraint (Gnocchi) | Chen et al., Nature 2024 |
| gnomAD Regional Missense Constraint | Wang et al. 2024; method Samocha et al. 2017 |
| Pangolin | Zeng & Li, Genome Biol 2022 |
| Repeat regions | Smit et al.; Benson, NAR 1999 |
| ClinGen | Rehm et al., NEJM 2015 |
| UniProt | UniProt Consortium, NAR 2025 |
| Reactome / GO | Milacic et al. 2024; Ashburner et al. 2000 |
| MaveDB | Esposito et al., Genome Biol 2019 |
| CIViC | Griffith et al., Nat Genet 2017 |
| Orphanet | Rath et al., 2012 |