General considerations
You can run predictions on OCHEM using simple REST-like web services.
| Info |
|---|
Some OCHEM models (e.g. models aggregated using Bagging method), require significant amount of calculations. |
Way 1: Request until done
To post a prediction task for a molecule, run the following request:
https://ochem.eu/modelservice/getPrediction.do?modelId=YOUR_MODEL_ID&mol=YOUR_MOLECULE
Model id the public model identifier or "model identity" (GUID) provided by OCHEM administration. If you plan to use predictions systematically, it is recommended to request a model identity (GUID).
The molecule can be in SMILES or SD format. For both the formats, multiple molecules can be posted using $$$$ separator.
It is much more efficient to predict molecules in batches rather than posting separate tasks for each molecule.
The resulting JSON will look like:
| Code Block | ||
|---|---|---|
| ||
{
"status" : "pending",
"taskId" : 0,
"metaserverTaskId" : 0
} |
The above response means that this molecule is new and it is being calculated at the moment.
Please, repeat the request at the periodic intervals (every 5-10 seconds) until the prediction result is returned in the following format:
| Code Block | ||
|---|---|---|
| ||
{
"taskId": "0",
// Array of predictions (for each input molecule)
"predictions": [{
"moleculeID": "1002136505",
// Array of predictions for a given molecule. Normally, contains only one prediction.
// Can contain multiple predictions for multi-models
"predictions": [{
"unit": "-log(mmol/L)",
// Name of the predicted class for classification models. Same as "value" for regression models.
"predictedValueString": "2.71",
"value": "2.71", // Prediction value (round it as you find necessary)
"dm": "0.86", // The "distance to model" used for the accuracy estimation
"insideAD" : true, // is this molecule inside the model's applicability domain?
"property": "log(IGC50-1)", // The predicted property
"accuracy": "0.70", // The prediction accuracy (RMSE)
"realValue": "0.0"
}],
"depictionID": "1000651576"
}],
"metaserverTaskId": "0",
"status": "success",
"modelDescriptionUrl": "http://ochem.eu/model/3"
} |
Way 2: Using task IDs
Since a prediction is not instantaneous and can take several seconds to minutes, the prediction is performed asynchronously, that is in two steps:
- Start a prediction task and get a task ID
- Fetch your prediction task using the task ID from step (1). Keep fetching until the task is ready
The API for these two simple steps is described below.
To post a task, run the following request:
https://ochem.eu/modelservice/postModel.do?modelId=YOUR_MODEL_ID&mol=YOUR_MOLECULE
The resulting JSON will look like:
| Code Block | ||
|---|---|---|
| ||
{
"taskId": "1000042989", // This is the task ID you need to know
"metaserverTaskId": "-1",
"status": "success"
} |
Given that the result.modelResponse.status is "success", the task ID used for retrieving the predictions is result.modelResponse.taskId
To fetch the model, use the following request:
http://ochem.eu/modelservice/fetchModel.do?taskId=YOUR_TASK_ID
If the task is still running, the resulting JSON will look like:
| Code Block |
|---|
{
"taskId": "0",
"metaserverTaskId": "0",
"status": "pending" // Keep requesting at periodic intervals, while the status is "pending"
} |
When the task is ready, the JSON will look like:
| Code Block |
|---|
{
"taskId": "0",
// Array of predictions (for each input molecule)
"predictions": [{
"moleculeID": "1002136505",
// Array of predictions for a given molecule. Normally, contains only one prediction.
// Can contain multiple predictions for multi-models
"predictions": [{
"unit": "-log(mmol/L)",
// Name of the predicted class for classification models. Same as "value" for regression models.
"predictedValueString": "2.71",
"value": "2.71", // Prediction value (round it as you find necessary)
"dm": "0.86", // The "distance to model" used for the accuracy estimation
"property": "log(IGC50-1)", // The predicted property
"accuracy": "0.70", // The prediction accuracy (RMSE)
"realValue": "0.0"
}],
"depictionID": "1000651576"
}],
"metaserverTaskId": "0",
"status": "success",
"modelDescriptionUrl": "http://ochem.eu/model/3"
} |
Way 3: Using single SMILES
Another and a simpler way to get prediction is to use
https://ochem.eu/modelservice/getPrediction.do?modelId=YOUR_MODEL_ID&mol=YOUR_MOLECULE
request, e.g.,
https://ochem.eu/modelservice/getPrediction.do?modelId=1&mol=c1ccccc1
to predict mutagenicity using AMES model https://ochem.eu/model/1
You will get a similar JSON results as above, but without a need to use explicitly taskId. This calculation, however, can be performed with only a single molecule.