Table 2 Summary of t-SNE optimizations proposed in opt-SNE workflow
Parameter | opt-SNE setup | Suggested use with cytometry data |
|---|---|---|
Gradient descent learning rate | Adaptive learning rate with initial value η = n/α, where n is the number of datapoints and α is the early exaggeration factor | Automated per dataset |
Early exaggeration factor | Standard t-SNE setup considerations apply | 4–12 |
Perplexity | Standard t-SNE setup considerations apply | 30–50 |
Early exaggeration termination | KLD value (cost function) is monitored in real time, and early exaggeration is removed at maxKLDRC | Automated per dataset |
t-SNE termination | KLD value (cost function) is monitored in real time and the embedding is finalized when (KLDN−1−KLDN) < KLDN/X | X = 5000 |
