Appendix 1—table 5. Computation time across different transfer learning methods under the same hardware conditions (128 GB RAM, 16 core processor).
| Wall time (Pancreas dataset) | |||
|---|---|---|---|
| Methods | Source Model (~12 K cells) | Target Model (~2 K cells) | Total time |
| ItClust | 2 min 4 s | ~2 min | |
| SCellBOW (thread = 16) | 2 min 5 s | 1 min 20 s | ~3 min |
| SCellBOW (thread = 1) | 6 min 21 s | 2 min 8 s | ~8 min |
| scETM (600 epoch, thread = 16) | 22 min 38 s | 5 min 46 s | ~27 min |
| scETM (600 epoch, thread = 1) | 23 min 49 s | 5 min 58 s | ~28 min |
| scBERT | 3 hrs 33 min | 2 min | ~3 hrs |
The term ‘thread’ represents the number of threads used: thread = 1 indicates single-threaded execution, while thread >1 indicates multi-threaded execution.