Neural DNA Alignment Demo
Neural DNA Alignment with Learned Embeddings
GenoCache uses learned 512bp embeddings + FAISS seeding to beat minimap2 on long noisy reads. See neural seeding in action with real HG002 ONT data.
Neural Seeding
GenoCache encoder trained on GRCh38 with 512bp context windows
FAISS Index
IVFPQ approximate nearest neighbor search over genome-wide embeddings
Real Metrics
See how seeds/read affects mapping coverage from 71% to 99%
Beats minimap2
97-99% mapped with 32+ seeds vs 94.5% minimap2 baseline
Neural Seeding Performance
GenoCache alignment results on HG002 ONT long reads across different seed configurations
| Configuration | Seeds | K | Mapped | vs minimap2 |
|---|---|---|---|---|
| GenoCache (Low) | 7→13 | 32 | 71% | -23.5% |
| GenoCache (16 seeds) | 16 | 32 | 84% | -10.5% |
| GenoCache (Normal) | 32 | 32 | 97% | +2.5% |
| GenoCache (High) | 64 | 32 | 99% | +4.5% |
| minimap2 baseline | - | - | 94.5% | - |
Measured on 100-read HG002 ONT subset. Neural seeding with 32+ seeds achieves higher coverage than minimap2's short hash-based seeds.
Ready to try GenoCache Align?
Run the demo with real HG002 ONT reads and see how neural seeding with learned embeddings can achieve higher mapping rates than traditional hash-based approaches.
Start Demo Run