I think, given the long-sequence dynamics of DNA, it would be worthwhile to try the SigFormer Architecture (instead of a pure Transformer architecture), combining Mamba and Transformer Architecture and keeping the computational costs manageable.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Exact numerical representation | 2 | 971 | September 1, 2025 | |
| Any way to access embeddings? | 5 | 1468 | April 2, 2026 | |
| Right syntax for dna_model.predict_sequence | 1 | 852 | August 8, 2025 | |
| How to visualize predictions made with predict_sequence? | 5 | 2346 | August 11, 2025 | |
| How could AlphaGenome predict Transcriptome from a Genome ❓ | 2 | 1042 | September 20, 2025 |