SteppeDNA — Multi-Gene HR Variant Classifier

Per-Gene ROC-AUC Performance

BRCA2

0.983

RAD51D

0.804

RAD51C

0.743

BRCA1

0.706

PALB2

0.641

Overall AUC (0.978) is weighted by gene representation. BRCA2 comprises 52% of the test set. Evaluated on a held-out 20% stratified test split.

Non-BRCA2 temporal AUCs (0.51–0.61) indicate limited generalization over time for data-scarce genes.

Per-Gene SOTA Comparison

Gene	SteppeDNA	Best SOTA	Winner
BRCA2	0.983	0.949 (BayesDel)	SteppeDNA
RAD51D	0.804	0.461 (REVEL)	SteppeDNA
RAD51C	0.743	0.703 (CADD)	SteppeDNA
BRCA1	0.706	0.646 (BayesDel)	SteppeDNA
PALB2	0.641	0.732 (REVEL)	REVEL
All scores evaluated on SteppeDNA's own test set, giving SteppeDNA a methodological advantage. Independent benchmark AUC: 0.719–0.793. REVEL, BayesDel, and CADD are general-purpose tools not trained on this distribution.

Evaluated on SteppeDNA’s held-out test set. Independent benchmark AUC: 0.719–0.793.

Upload VCF

Drop your VCF file here

or click to browse · .vcf format

Manual Entry

cDNA Position (1 - 10257)

Reference AA

Alternate AA

Mutation (optional)

Research Use Only — not for clinical diagnosis | Germline variants only

Pathogenic XGBoost · Calibrated · SHAP explained

Pathogenicity Probability 98%

Prediction based on AlphaMissense, MAVE, PhyloP, ESM-2, SpliceAI, AlphaFold 3D structure, and 103 engineered features.

VCF Batch Results

#	Type	HGVS	cDNA	AA Change	Mutation	Prediction	Probability

Universal Base Models

Multi-gene HR Deep Neural Network & XGBoost ensemble trained on 19,000+ ClinVar & gnomAD HR missense variants with isotonic probability calibration. Resolves 5 genes seamlessly.

103 Features

BLOSUM62 substitution scores, ESM-2 protein language model embeddings, PhyloP conservation, MAVE functional assays, AlphaMissense, SpliceAI, AlphaFold 3D structures, and gnomAD population frequencies.

SHAP Explanations

Every prediction shows which features pushed it toward pathogenic or benign, using SHAP values extracted from the XGBoost model.

Model Card & Limitations

Intended Use

Research-grade variant classification for missense mutations in 5 Homologous Recombination DNA repair genes (BRCA1, BRCA2, PALB2, RAD51C, RAD51D). Intended as a decision-support tool, not a standalone diagnostic.

Training Data

19,223 variants: 18,738 from ClinVar (reviewed/expert-panel classifications) + 485 gnomAD proxy-benign (AC ≥ 2 in ~1.6M alleles). 60/20/20 split with gene × label stratification. RANDOM_STATE=42.

Architecture

XGBoost (60%) + Multi-Layer Perceptron (40%) blended ensemble with isotonic calibration trained on a held-out calibration set. 103 engineered features from 5 databases.

Performance

ROC-AUC: 0.978 · MCC: 0.881 · Balanced Accuracy: 94.1%. 10-fold CV: 0.9797 ± 0.0031. Outperforms REVEL (0.725), BayesDel (0.721), CADD (0.539) on same test set.*

* Evaluated on SteppeDNA test set. General-purpose tools not trained on same distribution.

Known Limitations

Non-BRCA2 genes have smaller training sets and lower per-gene AUC (PALB2: 0.641, RAD51C: 0.743)
Copy number variants (CNVs) are not classified. Splice-site, in-frame indel, and synonymous variants use rule-based classification (not the ML model). Truncating variants (nonsense, frameshift) use Tier 1 protocol
Model may underperform on novel variants with no population frequency data
Probabilities are clipped to [0.5%, 99.5%] — the model never claims absolute certainty
Compound heterozygosity warning is provided in VCF batch analysis, but does not perform phasing; variants require segregation study to determine cis/trans configuration
Training data predominantly from populations of European descent (ClinVar submission bias); predictions for underrepresented ancestries may be less reliable
ClinVar classifications evolve over time; training labels may not reflect the latest expert consensus

Ethical Considerations

Not validated for clinical diagnostic use — all predictions require expert review
Training data is primarily from populations of European descent; performance may vary across ancestries
ACMG evidence codes are computational approximations, not expert classifications
Should not be used as the sole basis for clinical decisions about patient care

Regulatory & Certification Roadmap

Future Kazakhstan Ministry of Health Registration

Future CE-IVD Marking (EU)

Active Research Use Only (RUO)

No regulatory approval has been obtained or is currently in progress. The roadmap below reflects aspirational milestones only.

This tool is currently for research use only and has not been submitted for regulatory approval. Clinical deployment will require certification from the relevant national health authority.