Table 1:
GED, virtual screening, and timing results. Best is bold, second best is underlined. Runtime is proportional to the dimensionality, not method of dimensionality reduction, therefore methods with the same dimension have identical query times.
| Query accuracy | Query time (s) | ||||
|---|---|---|---|---|---|
|
| |||||
| Method | D | VS AUC | GED | Brute-force | k-d tree |
|
| |||||
| ECFC-256 | 256 | 0.494 | 6.69 | 24, 900 ± 2, 200 | N/A |
| ECFP-256 | 0.566 | 8.30 | 13,600 ± 1,000 | ||
|
| |||||
| SparseRP-16 | 0.501 | 15.06 | |||
| PCA-16 | 16 | 0.524 | 6.97 | 13,000 ± 70 | 37.5 ± 26 |
| SmallSA-16 | 0.692 | 5.74 | |||
|
| |||||
| PCA-8 | 8 | 0.517 | 8.78 | 11,400 ± 98 | 0.606 ± 0.35 |
| SmallSA-8 | 0.662 | 7.30* | |||