NAV Navbar
python shell
  • Introduction
  • Client access tokens
  • API Usage
  • Errors
  • Introduction

    Documentation for the Prediction API. This API is currently in ALPHA.

    The API is secured and you need credentials to be able to use it. We use the industry-standard OAuth2 protocol. For those of you who'd like more information about the ins and outs of OAuth2 please take a look here. Note that using the API is incredibly easy; you don't really need an in-depth understanding of OAuth2 to be able to use this API. This documentation contains all the information you need to get started!

    Client access tokens

    import requests
    data =
        headers={"Content-Type": "application/json"},
    # Inspect the response
    curl --request POST \
      --url \
      --header 'content-type: application/json' \
      --data '{"client_id":"<client_id>","client_secret":<client_secret>, "audience":"","grant_type":"client_credentials"}'

    Returns the following JSON response:

    { "access_token": "<access_token>", "expires_in": 86400, "token_type": "Bearer" }


    # When your client id is invalid
    { "error": "access_denied", "error_description": "Unauthorized" }

    We use the service to manage authentication and authorization. Please check that you have received a client_id and a client_secret before continuing. Retrieval of your access token is a two-step process:

    1. Your application authenticates with Auth0 using its client_id and client_secret.

    2. Auth0 validates this information and returns an access_token.

    Retrieving the access token for interaction with our API can be achieved by sending a request to the Auth0 authentication servers. Tokens have a lifespan of 24 hours. You need to re-acquire a new one before the old one runs out in order to keep using the API without interruption.

    We recommend the requests library for Python for ease of use.

    API Usage

    Having an access_token, you are now ready to use the API.

    Transcripts overview

    transcripts = requests.get(
        headers={"Authorization": "Bearer <access_token>"}
    curl --request GET \
      --url \
      --header 'Authorization: Bearer <access_token>'

    Returns the following JSON response:

    ["ENST00000257700", "ENST00000259008", "ENST00000260947", "ENST00000261584", ... ]

    Returns a full list of the transcripts available on our API. Each of these can be used in other calls. Transcripts are formatted as Ensemble identifiers. This list is retrieved from the /transcripts endpoint.

    Retrieving variant predictions

    All the request parameters are contained in the URL

    variant_data = requests.get(
        headers={"Authorization": "Bearer <access_token>"}
    curl --request GET \
      --header 'Authorization: Bearer <access_token>'

    A single endpoint for variant prediction retrieval exists: /transcripts/<transcriptId>/predictions/<residueNumber>/<variantType>. Using this endpoint requires the following:

    Returns the following JSON response:

        "residue_type": "Arg",
        "residue_number": 110,
        "variant_type": "Ala",
        "source_max": 0.3202865839,
        "source_min": 0.259611775,
        "spread": 0.0606748089,
        "transcript_id": "ENST00000622645",
        "deleterious": 0.3066707837,
        "effect_summary": "0.3067 (benign, high confidence)"
    Identifier Description
    residue_type Wildtype amino acid
    residue_number 1-based protein sequence number
    variant_type Variant amino acid type
    source_max Highest sub-predictor prediction
    source_min Lowest sub-predictor prediction
    spread Difference between highest and lowest sub-predictions. Can be interpreted as a measure of uncertainty.
    transcript_id Ensembl transcript id
    deleterious Value indicating deleteriousness. Value between 0 and 1, where a value between 0 and 0.5 means benign and a value between 0.5 and 1 means deleterious.
    effect_summary A simple summary of the prediction with assesment of the impact on the protein and the confidence in the prediction. The first is based on the 'deleterious' score, the second is based on the 'spread' score. Our spread indicates the distance between predictions of our subclassifiers in our ensemble. If the score is below 0.15, we determine that the results of the subclassifiers are concordant, resulting in "high confidence". Between 0.15 and 0.25 indicates "medium confidence" and scores above 0.25 indicate "low confidence".


    Currently, all errors (including authentication errors) will yield a 500 response and a JSON body, containing {"Error": "Internal Server Error"}.