"The same CRISPR guide safe for one patient can be dangerous for another — because their genomes are different."
-purple){kind=icon}
Flagship case — Parkinson's disease (GBA1 / N370S): 6 CRISPR guides analyzed against the full human genome, revealing 3 HIGH-risk off-targets hidden inside the GBAP1 pseudogene — the exact risk a naive, gene-only design would miss. Therapeutic mRNA scored 96/100.
Parkinson's disease affects ~10 million people worldwide, and — as of July 2026 — there is no approved disease-modifying therapy. Every treatment on the market only manages symptoms; none corrects the underlying biology.
The single largest genetic risk factor for Parkinson's is the GBA1 gene (glucocerebrosidase / GCasa), implicated in ~5–15% of cases:
Gene: GBA1 (Glucocerebrosidase / GCasa) — Chromosome 1
Variant: c.1226A>G | rs76763715 | N370S (p.Asn409Ser)
Codon: AAC → AGC
Protein: Asparagine → Serine → reduced lysosomal GCasa activity
Result: Glucosylceramide accumulation → elevated Parkinson's risk
Status: NO approved disease-modifying therapy (Jul 2026); GBA1 approaches in trials
GBA1 sits ~16 kb away from a pseudogene, GBAP1, with ~96% sequence identity. Any CRISPR guide designed for GBA1 looks almost identical to regions of GBAP1 — a natural minefield of near-identical off-targets. A guide that appears "clean" against the gene alone can, in reality, cut the pseudogene.
This is exactly where Exonik's engine — real off-target analysis against the whole genome, plus patient-level SNP personalization — delivers its differential value.
Sickle Cell Disease (SCD) is not forgotten — it remains Exonik's fully validated baseline (see below). But Parkinson-GBA1 is where the platform proves it can attack a high-impact disease with no approved cure and genetically hostile terrain.
A conventional "design a guide on the gene sequence" approach would ship guides GBA1-g86/g87/g88 as reasonable candidates. Exonik ran each guide against the entire GRCh38 reference genome and found the hidden danger:
Guide GBA1-g86 / g87 / g88:
ON-target: chr1:155,235,xxx (GBA1) ✅ 0 mismatches
OFF-target: chr1:155,215,xxx (GBAP1 pseudogene) ⚠️ HIGH RISK — 0 mismatches!
→ A perfect-match cut in the WRONG place.
Guide GBA1-g240 / g85 / g241:
ON-target only. 0 off-targets. Safety score 100/100 ✅
3 guides flagged HIGH-risk (perfect-match cut inside GBAP1) — and Exonik proposes 3 alternative guides with 100/100 safety. This is the difference between a therapy and an accident.
Nobody is doing genome-wide, patient-personalized off-target screening at scale for these targets. That is Exonik's opportunity.
Exonik is disease-agnostic by design. The core logic (CRISPR design, off-target BLAST, mRNA design, dashboards) is shared; each disease is a self-contained module (diseases/<name>.py) defined by a standard Disease schema. Switching targets is a single environment variable:
# Run the whole pipeline for a different disease — same engine, no code changes
export EXONIK_DISEASE=parkinson_gba1 # or: sca | beta_thal| Disease | Gene | Variant | Role in Exonik |
|---|---|---|---|
| Parkinson (GBA1) | GBA1 | N370S · rs76763715 | ⭐ Flagship — unmet need + GBAP1 differentiator |
| Sickle Cell Disease | HBB | rs334 · GAG→GTG | ✅ Validated baseline — 5 real patients, full personalization |
| Beta-Thalassemia | HBB | (β⁺/β⁰) | 🧩 Proof of modularity — same gene, different variant |
The same platform that was fully validated end-to-end on Sickle Cell was redirected to Parkinson-GBA1 by configuration alone — the strongest possible evidence that Exonik scales across monogenic and genetically complex targets.
Google DeepMind's Isomorphic Labs is applying AlphaFold and generative AI to revolutionize drug discovery — designing small molecules that bind to protein targets. Their framework is impressive:
Isomorphic Labs (DeepMind):
Protein Structure (AlphaFold) → Drug Target → Small Molecule Design → Clinical Trials
= Find the broken protein, design a molecule that patches it
But Exonik takes a fundamentally different approach:
EXONIK:
Patient Genome → Mutation Identified → CRISPR Guide → mRNA Therapy → Gene Corrected/Restored
= Find the broken DNA, fix (or restore) the code that builds the protein
| Isomorphic Labs (Drug Discovery) | EXONIK (Gene Therapy) | |
|---|---|---|
| Target | Protein surface (downstream) | DNA / mRNA (upstream) |
| Action | Design molecule to bind/block protein | Edit genome / restore the missing protein |
| Duration | Chronic (patient takes drug repeatedly) | One-time correction / durable restoration |
| Personalization | Same drug for all patients | Each patient's genome analyzed |
| Analogy | Patching a bug at runtime | Fixing the source code |
Both approaches use AI + structural biology. But while drug discovery treats the symptom (a misfolded protein), gene therapy fixes the cause (a mutated or deficient gene). Exonik operates at the most upstream point possible — the genome itself.
"The best way to fix a bug is not to write a better error handler — it's to fix the line of code that causes it."
EXONIK is a computational pipeline that designs precision, patient-personalizable gene therapies — from raw genomic data to clinical report — fully automated.
It combines:
- CRISPR/Cas9 guide design with real off-target analysis against the full human genome (GRCh38)
- Genome-wide safety screening that catches near-identical off-targets (e.g. the GBAP1 pseudogene)
- Patient-level personalization — each patient's SNPs crossed against every off-target (validated on SCA; roadmap for Parkinson)
- Therapeutic mRNA design — codon optimization, secondary structure prediction, immunogenicity screening
- AlphaFold 3D visualization — protein structure analysis with CRISPR GPS mapping
- Interactive clinical dashboard — ready for presentation to clinicians or investors
Two layers of intelligence: precision at the variant (design for the disease-causing mutation) + personalization at the patient (adapt to each individual genome).
| Metric | Result |
|---|---|
| Target gene | GBA1 (chr1) — N370S / rs76763715 |
| Approved disease-modifying therapy | None (as of Jul 2026) |
| CRISPR guides designed | 6 (analyzed vs full GRCh38 genome) |
| Guides with perfect safety | 3 (GBA1-g240, g85, g241) — 0 off-targets, 100/100 |
| HIGH-risk off-targets detected | 3 — all inside the GBAP1 pseudogene (0-mismatch cuts) |
| Pseudogene challenge | GBAP1 ~96% identity, ~16 kb from GBA1 |
| Therapeutic mRNA score | 96/100 (functional GCasa, 536 aa) |
| AUG accessibility | 100/100 |
| Immunogenicity risk | VERY LOW (95/100 — endogenous protein) |
| Suggested delivery | LNP → IV, blood-brain-barrier-crossing ligands |
| Metric | Result |
|---|---|
| Real patients analyzed | 5 (Nigeria, Gambia, Colombia, Puerto Rico, Caribbean) |
| CRISPR guides designed | 12 (6 HBB + 6 BCL11A strategies) |
| Off-targets found (real BLAST) | 12 (chr2, chr11, chr14) |
| Personalization coverage | 12/12 off-targets × 5 patients |
| Best guide | HBB-g68 (safety: 100/100, 0 off-targets) |
| Therapeutic mRNA score | 96/100 |
| 3D visualizations | 4 interactive HTML + 3 static diagrams |
Open
reporte_integrado/parkinson_gba1/dashboard_exonik.htmlin any browser — no server required.
Executive summary + the unmet-need / differential-value narrative: 6 guides, 9 real off-targets, 3 HIGH-risk (GBAP1), 3 perfect guides, mRNA 96/100.
Per-guide safety scores and the exact GBAP1 off-targets — green guides are ready, red guides are the ones a gene-only design would have shipped by mistake.
The flagship dashboard includes:
- Executive Summary — headline metrics at a glance
- Differential Value & Unmet Need — why this matters and what Exonik uniquely detects
- CRISPR Safety Profile — safety scores for all guides (GBA1 vs GBAP1)
- Genomic Off-Target Map — where each hit lands in the genome
- mRNA Design — multidimensional therapeutic quality profile
The Sickle Cell dashboard (
reporte_integrado/dashboard_exonik.html) adds a Patient × Guide personalization heatmap across the 5-patient cohort — the personalization layer that is on the roadmap for Parkinson.
Exonik maps CRISPR guides from DNA coordinates all the way to the 3D protein structure (AlphaFold / RCSB PDB) — visualizing exactly where the guide acts on the protein. This capability is fully demonstrated on Sickle Cell (HBB):
Left: HBB protein with 8 alpha helices colored — Right: GPS CRISPR showing exactly where guide HBB-g68 cuts at GLU-7 (the SCA mutation site).
Guide RNA: 5'-TAACGGCAGACTTCTCCTCA-3' (antisense)
CDS positions: 17 → 36 (20 nt) | Protein zone: amino acids 6 → 12
Cas9 cut site: inside codon 7 = GLU-7 = rs334 mutation
Safety score: 100/100 | Off-targets: 0
| Visualization | File | Description |
|---|---|---|
| HBB Spectrum | estructuras_3d/HBB_3d_espectro.html |
Full protein colored N→C terminal |
| Mutation Site | estructuras_3d/HBB_3d_mutacion.html |
GLU-7 highlighted in Helix A |
| 8 Alpha Helices | estructuras_3d/HBB_3d_helices.html |
All helices A–H color-coded |
| GPS CRISPR | estructuras_3d/GPS_CRISPR_3d.html |
HBB-g68 target zone on 3D protein |
🗺️ Roadmap — GBA1 3D: The same GPS-CRISPR view for GCasa (GBA1, UniProt P04062) with the N370S site highlighted is the next visualization. Because 3D operates at the protein level (not patient data), it applies directly to Parkinson.
| Tool | Version | Purpose |
|---|---|---|
| Python | 3.10+ | Core pipeline language |
| BLAST+ | 2.17.0 | Off-target search against full genome |
| seqfold | latest | mRNA secondary structure (thermodynamic) |
| pyliftover | latest | GRCh38 ↔ GRCh37 coordinate conversion |
| plotly | 5.x | Interactive clinical dashboard |
| py3Dmol | 2.x | Interactive 3D protein visualization |
| BioPython | 1.8x | PDB file parsing and analysis |
| AlphaFold DB / RCSB PDB | — | Protein 3D structures |
| 1000 Genomes Project | Phase 3 | Real genomic variants from 2,504 people |
| GRCh38.p14 | NCBI | Human reference genome (~3.1 GB) |
exonik/
├── config.py ← Dynamic disease loader (EXONIK_DISEASE)
├── fase_01_genomas_reales.py ← Phase 1: Download & parse 1000 Genomes data
├── fase_05_reporte_integrado.py ← Phase 5: Generate dashboards & clinical reports
├── requirements.txt ← Python dependencies
├── README.md ← This file
│
├── diseases/ ← MODULAR DISEASE DEFINITIONS (public reference data)
│ ├── base.py ← `Disease` dataclass — the shared schema
│ ├── sca.py ← Sickle Cell (HBB) — validated baseline
│ ├── beta_thal.py ← Beta-Thalassemia (HBB) — modularity proof
│ └── parkinson_gba1.py ← Parkinson (GBA1) — flagship
│
├── reporte_integrado/ ← DEMO OUTPUTS (open in browser)
│ ├── dashboard_exonik.html ← SCA dashboard (personalization heatmap)
│ ├── reporte_clinico_HG01889.html ← SCA sample clinical report
│ └── parkinson_gba1/
│ └── dashboard_exonik.html ← Parkinson flagship dashboard
│
├── estructuras_3d/ ← 3D INTERACTIVE VISUALIZATIONS (HBB)
│ ├── HBB_3d_espectro.html / _mutacion.html / _helices.html
│ └── GPS_CRISPR_3d.html ← CRISPR guide mapped to 3D protein
│
└── screenshots/ ← STATIC DIAGRAMS & SCREENSHOTS
├── Parkinson_Dashboard_Resumen.png ← Flagship: summary + differential value
├── Parkinson_Dashboard_CRISPR.png ← Flagship: GBA1 vs GBAP1 guide safety
├── HBB_VS_Crispr.png ← SCA 3D: protein helices vs CRISPR GPS
├── Guia_CRISPR_Dashboard.png ← SCA dashboard: 12 CRISPR guides
└── ARNm_Dashboard.png ← SCA dashboard: mRNA therapeutic design
Note: Phases 2, 3, 4 and the 3D notebook (Phase 6) contain proprietary algorithms (core IP) and are available to research partners and collaborators under NDA. The
diseases/modules contain only public reference sequences (NCBI / dbSNP) and the modular architecture — intentionally shown to demonstrate the platform design.
Phase 2 → 6 CRISPR guides designed on GBA1; each BLASTed vs the full GRCh38 genome
↓
→ 3 guides score 100/100 (zero off-targets) — clinically usable
→ 3 guides flagged: HIGH-risk 0-mismatch cut inside the GBAP1 pseudogene
↓
Phase 4 → Therapeutic mRNA for functional GCasa designed (96/100), LNP + BBB delivery
↓
Phase 5 → Flagship dashboard: unmet need + GBAP1 differentiator + mRNA evidence
↓
Phase 3 (roadmap) → Cross the GBAP1 off-targets with each patient's real SNPs
→ "Is this guide safe for THIS patient's genome?"
From genome to atom. From data to therapy. Precision at the variant — personalization at the patient.
Research Use Only (RUO) This software is intended for research and educational purposes only. It is NOT approved for clinical use, diagnosis, or treatment. All patient identifiers are derived from public research cohorts (1000 Genomes Project) and are used in accordance with their open data access policy. Reference sequences for all disease modules are public (NCBI / dbSNP). Any application of these results to real patients requires independent laboratory validation and regulatory approval.
Parkinson's disease affects ~10 million people and still has no approved disease-modifying therapy. Exonik is a computational engine that designs precision, personalizable gene therapies — reducing design time from months to hours — and, critically, catches the hidden risks (like the GBAP1 pseudogene) that a gene-only approach would miss. Validated end-to-end on Sickle Cell, redirected to Parkinson by configuration alone.
Exonik — Precision & Personalized Gene Therapy Design Platform Built with Python, BLAST+, AlphaFold, real genomic data, and a lot of curiosity about programming cells.
Version: 0.2.0 (Prototype — Modular Platform) | Last updated: July 2026