Retrieving predictions
Oktoberfest relies on retrieving predictions from a Koina server that hosts specific models for peptide property prediction. Users can use any publicly available community server or host their own server.
Connecting to a community server
Intensity models |
Description |
|---|---|
Prosit_2019_intensity |
Developed for HCD tryptic peptides only. We recommend using the Prosit_2020_intensity_HCD model instead, since it showed slightly superior performance on tryptic peptides as well. |
Developed for HCD tryptic and non-tryptic peptides. Supported modifications are oxidation and carbamidomethylation. Latest version we recommend to use for HCD. |
|
Developed for CID tryptic and non-tryptic peptides. Supported modifications are oxidation and carbamidomethylation. Latest version we recommend to use for CID. |
|
Developed for HCD and CID, tryptic and non-tryptic peptides. Latest version we commend for TMT labeled peptides in general. |
|
Developed for timsTOF, tryptic and non-tryptic peptides. Latest version we commend to use for timsTOF. |
|
Developed for HCD cleavable cross-linked peptides (DSSO, DSBU). Supports oxidation and carbamidomethylation. |
|
Developed for HCD non-cleavable cross-linked peptides (DSS, BS3). Supports oxidation and carbamidomethylation. |
|
Developed for generic data support, including TMT, timsTOF and various instrument types. |
|
Developed for HCD tryptic and non-tryptic peptides. |
iRT models |
Description |
|---|---|
While developed for tryptic peptides only, we did not observe a drop in prediction performance for non-tryptic peptides. |
|
Developed for TMT labeled peptides. |
|
Developed for for generic data support, including TMT, timsTOF and various instrument types. |
Once support for additional models is implemented in Oktoberfest, they will be added here.
Hosting and adding your own models
In case you are planning to host your own private or public instance of Koina or want us to host your model, please refer to the official Koina documentation.
Local intensity prediction & refinement learning
Instead of using a pre-trained intensity predictor via Koina, you can also predict intensity locally from a DLomix model. If you do not have a pre-trained intensity predictor in .keras format, a baseline model will automatically be downloaded for you. Importantly, this also gives you the option to refinement-learn the pre-trained predictor on your input dataset and using the refined model for intensity prediction.
For local intensity prediction and refinement learning, you need to provide either a path to a pre-trained model or the keyword baseline (for the runner to automatically download a model for you) as the intensity model in the config, and specify localPredictionOptions as well as optional refinementLearningOptions. For more details, refer to the job and configuration docs.