XGBoost · SHAP · Gemini AI · v2.0

Clinical Genomic
Variant Intelligence

Instantly classify pathogenic vs. benign variants with SOTA XGBoost trained on 300K ClinVar records — augmented with SHAP explainability and Gemini AI clinical notes.

🔬 Try Free Demo View Pricing →

300K+

ClinVar Variants

97.2%

AUC Score

<200ms

Inference Latency

SHAP Features

Platform Features

Built for Clinical-Grade
Genomic Analysis

Every feature is designed with genomicists, clinical researchers, and bioinformaticians in mind.

⚡

SOTA XGBoost Model

Trained on 300K balanced ClinVar variants with categorical encoding for chromosome, REF, and ALT alleles. Best-in-class AUC of 97.2%.

XGBoost 2.0

📊

SHAP Explainability

Every prediction comes with full SHAP feature impact analysis — waterfall charts showing exactly which conservation scores drove the classification.

TreeExplainer

✨

Gemini AI Clinical Notes

Automatically generated clinical narrative explaining pathogenicity in plain language — biological context, feature breakdown, and actionable tips.

Gemini 2.5 Flash

🌍

Population Frequency (gnomAD)

Allele frequency from gnomAD integrated as a primary predictor. Rare variants flagged with higher pathogenic prior based on population data.

gnomAD v4

🧬

Conservation Scores

GERP++, phyloP100, phyloP470, phastCons470, phastCons17 — five evolutionary conservation metrics covering vertebrates, mammals, and primates.

5 Metrics

🔌

FastAPI REST Endpoint

Production-ready REST API with CORS support. Integrate directly into your LIMS, EHR pipeline, or research workflow via a simple POST request.

FastAPI · OpenAPI

How It Works

From Variant Input
to Clinical Insight

Submit Variant

Enter chromosome, position, REF/ALT alleles, gnomAD AF, and conservation scores via the UI or API.

XGBoost Inference

The SOTA model processes all 10 features with categorical encoding and predicts pathogenic probability in under 200ms.

SHAP Analysis

TreeExplainer computes per-feature SHAP values, revealing which conservation score or allele drove the prediction.

Gemini Clinical Note

Gemini AI synthesises a structured clinical note — verdict, feature breakdown, biological context, and practical tips.

Pricing

Simple, Transparent Pricing

Whether you're a solo researcher or a clinical diagnostics lab, there's a plan scaled to your needs.

Monthly

Yearly Save 20%

Starter

^$0 / mo

For students and researchers exploring the platform.

50 variant predictions / month
SHAP feature analysis included
Gemini AI clinical notes
GRCh38 / hg19 coordinate input
API access
Batch processing
Team workspace
Priority support

Try Free Demo

Trusted by Genomics Experts

★★★★★

The SHAP visualisation alone is worth the subscription. I can now explain to a clinician in 30 seconds why the model flagged a variant — that's game-changing for variant of uncertain significance cases.

Dr. Riya Mehta

Clinical Geneticist · AIIMS Delhi

★★★★★

We run ~2,000 WES samples monthly. PathoScan's batch API cut our variant triage time by 60%. The Gemini clinical notes are surprisingly accurate and save hours of manual reporting.

Sam Whitfield

Lead Bioinformatician · GenPath UK

★★★★☆

As a PhD student working on rare disease genomics, the free tier gives me more than enough for research. The 97% AUC is impressive — benchmarked it myself against in-house CADD scores.

Komal Nair

PhD Candidate · IISc Bangalore

FAQ

Frequently Asked Questions

Everything you need to know about PathoScan. Can't find the answer? Email us.

What training data does the XGBoost model use? ▾

The model is trained on 300,000 balanced ClinVar variants (Pathogenic vs. Benign) using 10 features including gnomAD allele frequency and five evolutionary conservation scores from dbNSFP. Data is preprocessed with rank normalisation for numerical stability.

How is the 97.2% AUC achieved? ▾

Through careful feature engineering (rank-score normalisation), categorical encoding of chromosome and alleles, class balancing on the training set, and hyperparameter tuning via Bayesian optimisation. The model is the same architecture (XGBClassifier with enable_categorical=True) as used in our research paper.

Is this a replacement for clinical variant interpretation? ▾

No. PathoScan is a decision-support tool and should complement — not replace — manual ACMG/AMP guideline-based classification by a certified clinical variant scientist. Results must be reviewed by a qualified professional before any clinical action.

How do I access the API? ▾

Pro and Enterprise subscribers receive an API key and access to our FastAPI endpoint at api.pathoscan.ai/v2/predict. Full OpenAPI 3.0 docs are available at api.pathoscan.ai/docs. Rate limits are 10 req/s on Pro, unlimited on Enterprise.

Can I run PathoScan on-premise? ▾

Yes — Enterprise plans include Docker/Kubernetes deployment packages, model weights, and on-site installation support. This is ideal for HIPAA-compliant environments where genomic data cannot leave your infrastructure.

Live Demo

Variant Pathogenicity Classifier

Predict whether a genomic variant is Pathogenic or Benign — powered by XGBoost trained on 300K balanced ClinVar variants.

Input Variant Parameters

Chromosome

Position (GRCh38)

REF Allele

ALT Allele

gnomAD Allele Frequency

GERP++

phyloP 100

phyloP 470

phastCons 470

phastCons 17

INFERENCE INITIALIZING...

SHA: —

—

Pathogenic probability: 0.00%

SHAP Feature Impact Analysis

← Benign PushPathogenic Push →

✦ Clinical Report:

Waiting for analysis...

⚠ For research and decision-support use only. Not for standalone clinical diagnosis.

Clinical Genomic Variant Intelligence

Built for Clinical-GradeGenomic Analysis

From Variant Inputto Clinical Insight