To obtain the most accurate results please use the MaGS predictor database. Query your protein of interest and if you obtain a Z-score greater than or equal to 1.16 for human or 1.08 in yeast then the protein is showing a high propensity to be included within phase-separated organelles within the cell. If the score is between 1.16 and 0.66 for human and 1.08 and 0.58 for yeast, then the protein has a modest likelihood, and biological factors other than the features used in the model could heavily influence condensate localization. If your score is below -0.36 for human or -0.39 for yeast, then the protein is showing a low likelihood for inclusion into membraneless organelles.
If MaGS does not provide a prediction score, likely due to the protein lacking experimental features needed for prediction by MaGS, or you have an unknown protein sequence then use the sequence-based predictor MaGSeq. If you obtain a MaGSeq value greater than or equal to 0.90 for human or 0.89 in yeast then the protein is highly likely part of phase-separated organelles within the cell. Similarly, scores between 0.90 and 0.56 for human and 0.89 and 0.25 for yeast are modest and scores below -0.33 for human and -0.45 for yeast are indicative of a low propensity result.
MaGSeq is intended to extend the scope of the MaGS method framework, however, scores are not directly comparable and MaGSeq should be used only when MaGS values are not available. Keep in mind that in order to obtain MaGSeq values, queries must have a minimal sequence length of 150 amino acids.
| Suggested Z-Score Cutoff Values | |||
|---|---|---|---|
| Predictor | High | Moderate | Low |
| MaGS Human | >1.16 | 1.16 to 0.66 | 0.66 to -0.36 |
| MaGS Yeast | >1.08 | 1.08 to 0.58 | 0.58 to -0.39 |
| MaGSeq Human | >0.90 | 0.90 to 0.56 | 0.56 to -0.33 |
| MaGSeq Yeast | >0.89 | 0.89 to 0.25 | 0.25 to -0.45 |
Please note, cutoff values are suggestions based on estimates of model worthiness from Receiver Operator Characteristic (ROC) Curves and model specificity at balance point. If you are performing a screen and desire a higher sensitivity, you would want to lower this threshold. Conversely, if you want to have a very high level of specificity in identifying granule proteins, then you may wish to raise the threshold. Furthermore, while MaGS and MaGSeq have been trained on stress granule proteins only, our analyses revealed that proteins localizing to other cellular granules have, on average, also increased prediction scores, probably because other membraneless organelles follow a similar mechanism of formation and, therefore, their constituents have related protein features. Thus, high z-scores are indicative of a protein’s capability to incorporate into a biological granule in the cell, but they do not indicate type. End users must determine, based on their own biological knowledge and experiments, which type of biological condensate is most likely. One of the first steps to refine this prediction is by examining cellular localization.
For more information please visit the documentation or about pages. If you need further assistance, please contact us directly. Also, if you have used these tools or found them helpful, please cite the following manuscripts: