Fig. 01 · DNA double helix · three.jsacquiring
+strand
ATGCTAGCGATCGATCGTAGCTAGCATCGGATCCATGCATGCTAGCGATCGATCGTAGCTAGCATCGGATCCATGCATGCTAGCGATCGATCGTAGCTAGCATCGGATCCATGCATGCTA
Start an analysis
Drag & drop FASTA or click to browse
.fasta, .fa, .fna · max 50 MB
or try a demo
Demo Samples
Why GATCHA — Significance & Scope
A defensive research prototype exploring trustworthy AI for antimicrobial resistance decision support.
Antimicrobial resistance could cause up to 10 million deaths per year by 2050, and lab testing takes 24–72 hours. GATCHA turns a genome into a fast, evidence-anchored second opinion.
Significance
- Faster than culture-based AST when minutes matter
- Calibrated probabilities — not opaque scores
- Every call is traceable back to a genomic marker or rule
- Honest no-call instead of confident guesses
In scope
- E. coli only — a well-studied model organism
- Ciprofloxacin, Gentamicin, Cefotaxime
- Assembled FASTA input (quality-checked)
- Ensemble: curated rules + logistic + deep model
Explicitly out of scope
- No treatment or dosing recommendations
- No other species or antibiotics
- No organism design or sequence modification
- Never a replacement for laboratory AST
A safety net, not a substitute. Outputs must be confirmed by standard laboratory antimicrobial susceptibility testing and reviewed by a qualified professional. GATCHA never prescribes, doses, or replaces AST.
Model Outcomes Comparison
How the rules layer, calibrated logistic model, and deep model each vote — and how the ensemble resolves them into the final call.
| Sample | Antibiotic | rules | Logistic P(R) | Deep P(R) | agreement | Final decision | Confidence |
|---|---|---|---|---|---|---|---|
| Loading real demo predictions… | |||||||
agree — models concur, high trust
split — logistic vs deep disagree
conflict — rules override triggered