Clopidogrel
CYP2C19
Poor metabolizer prevalence varies sharply across populations, which can change antiplatelet response and event risk.
PM prevalence: ~2% in many European cohorts vs much higher in parts of East Asia/Oceania.
Anukriti AI Private Limited
Anukriti AI
Modeling drug response across every ancestry — before the trial, not after.
Pharmacogenomic evidence skews European. We turn population structure into an auditable input — surfacing population-specific risk before protocol lock.
emoji_eventsAWS AIdeas Top 20
01Live Engine
Deterministic PGx, running now
Run a CYP2C19 simulation inline. Every call is auditable — CPIC + PharmVar provenance, named-rule refusals, and population-stratified output.
PGx TerminalDiagnostic Terminal — LOCALHOST:9000
settings
>
ANUKRITI AI PLATFORM INITIALIZED...
>
LOADING POPULATION-AWARE WORKFLOWS...
>
DETERMINISTIC PGX ENGINE READY
>
Target IdentifiedAX-4022 [Synthetic Phenotype]
In SilicoC12H18N2O2
Structure ID: AX-4022
3D-MODEL LIVE
Live Analysis Buffer
Workflows Live3 shipped
1000G Samples2,504 phase-3
Call Statuses3 states
Evidence LayerV1–V10 / R1–R12
Export ContractTrial-ready rows
Reference Cohorts
AFR EUR EAS SAS AMR
Global Mix
Decision Path
Deterministic
Auditable
Reviewer Flow
< 100ms warm
End-to-end
PM Frequency
CYP2C19Same drug, ~5× risk difference across ancestries · 1000G phase-3, n=2,504
02Platform Approach
Why Anukriti AI, why this approach
We focus on deterministic pharmacogenomic interpretation first, then layer AI explanations for accessibility. This architecture enables auditable decision pathways while helping teams explore population-level risk differences before costly trial-stage changes.
Validation Snapshot
IWPC warfarin cohort · n=5,700
low → 45.80→ standard → 33.66→ high → 21.58 mg/wk
Engine-predicted risk tier vs. actual prescribed dose on a public retrospective cohort. Monotonic gradient, ~24 mg/wk delta. A dose gradient is not a clinical outcome: this shows the engine's tiers track real prescribing, not that any patient was dosed wrongly.
Powered by Industry Standard Architecture
🧬1000 Genomes
🧪ChEMBL
⚙️RDKit
📚CPIC / PharmVar
☁️Bedrock / Nova
Recognition & Community Validation
AWS AIdeas Winner (Top 20)
Anukriti AI was selected as 1 of 20 winners from 10,000+ ideas submitted across 115 countries and featured on AWS Builder Center for its population-aware pharmacogenomics simulation approach.
shield_check
Research Simulation Guardrails
Anukriti AI is a research and education platform, not a clinical diagnostic system. Deterministic outputs remain auditable, AI-generated text is explanatory, and users should not treat simulation output as medical advice.
Now Live
What's shipped today
science
3 workflows
Clopidogrel/CYP2C19, Warfarin/CYP2C9+VKORC1, Simvastatin/SLCO1B1 — all CPIC level A, all auditable.
diversity_3
2,504 real samples
1000 Genomes phase-3 cohort across 5 superpopulations. Pre-resolved genotypes from AWS Open Data.
rule
Named-rule refusals
Evidence Sufficiency Layer: 12 R-rules, 10 V-rules, 9 U-rules. Every refused recommendation cites the rule id.
fact_check
IWPC cohort · n=5,700
Engine produces monotonic dose gradients (low 45.8 → high 21.6 mg/wk) on the public IWPC warfarin cohort — a retrospective consistency check, not a clinical validation, and no patient outcome is claimed from it. Companion CPIC table audit ships with open findings disclosed, including a CYP2C9 table mismatch found by our own audit.
Validation deep dive03Case Studies & Evidence
Why population-aware PGx matters
Anukriti AI is built on publicly available pharmacogenomics evidence and real-world case studies that show how genotype frequency differences can materially affect drug safety and efficacy outcomes.
biotech
Source: NCBI Medical Genetics Summaryopen_in_new
Warfarin
CYP2C9 + VKORC1
Dose requirements are strongly affected by PGx variation and can differ significantly across ancestry groups.
Multi-ethnic cohorts report meaningful dose separation linked to CYP2C9 and VKORC1 variants.
Carbamazepine
HLA-B*15:02
Severe hypersensitivity risk clusters in specific populations, illustrating why representation matters in early safety planning.
Strong association of HLA-B*15:02 with SJS/TEN in several Asian cohorts.
Codeine
CYP2D6
Ultra-rapid and poor metabolizer phenotypes can both create safety or efficacy failures depending on genotype distribution.
Marked global variability in CYP2D6 phenotype frequencies.
Evidence Links
- AWS Builder Center feature (Anukriti AI)
- Anukriti AI demo walkthrough
- UK pharmacovigilance PGx ADR analysis
- Global ancestry representation gap in GWAS
Research simulation only. Not intended for diagnosis, treatment, or prescribing decisions.
04Roadmap
From research prototype to trial intelligence
Four phases — deterministic core, reviewer trust layer, pilot operations, clinical-grade validation.
Phase 1Completed
Deterministic Trial Wedge
Three shipped workflows — clopidogrel (CYP2C19), warfarin (CYP2C9 + VKORC1 + CYP4F2), and simvastatin (SLCO1B1) — with auditable call statuses (called / cannot_call / insufficient_data) and CPIC + PharmVar provenance.
Live
Phase 2Completed
Reviewer Trust Layer
Benchmarks, confidence calibration, reproducibility packs, evidence mapping, reviewer reports. Evidence Sufficiency Layer with named refusals: 12 R-rules (decision), 10 V-rules (verdict), 9 U-rules (uncertainty), and 3 bias-detection kinds — every refusal cites a specific rule id.
Live
Phase 3In Progress
Pilot Operations
Production API on Azure Container Apps + MongoDB Atlas. Public simulation surface (zero-auth demo). Real 1000 Genomes phase-3 sample mode — pre-resolved genotypes from the AWS Open Data bucket. Per-cohort outcome distributions and shareable run permalinks.
Live
Phase 4In Progress
Clinical-Grade Validation
PharmCAT concordance benchmark, BCHE / Vysya founder-community wedge (Kerdoncuff 2025), CYP2D6 CNV calling via Cyrius, full AFR-specific evidence pack. Scientific partnership outreach (PharmVar, LASI-DAD, GenomeIndia) underway.
Next