Prophet Studio

A labeling tool for building your own domain knowledgebase from biomedical publications. Let us read the papers routinely and build our own domain knowledgebase together.

NOTE: If you only want to use the prophet studio but don't want to deploy it by yourself, you can access the online version of the prophet studio by https://prophet-studio.3steps.cn. The online version of the prophet studio is free for everyone, but you need to contact us to join an organization or register an organization first. After that, you can create a project and start labeling the papers. Please access the OpenProphetLab website to know more about our plan and contact us.

Why Prophet Studio?

In a researcher's daily work, he/she needs to read a lot of papers and extract the findings/knowledges from the papers into his/her mind. But it is hard to remember all the findings/knowledges from the papers. So we need a tool to help us to build our own domain knowledgebase from the papers. The Prophet Studio is one of answers for this problem.

In a prophet studio instance, you can fetch all the papers about your interested topics from the pubmed and label them with the label studio. After that, you can use the knowledge graph editor to annotate the findings/knowledges from the labeled papers and synchronize them into a biomedgps instance by manually or automatically. Finally, you can use the biomedgps to explore the knowledge graph and get a full knowledge graph about your interested topics. Like the following figure:

Components

A prophet studio instance is composed of the following systems:

Label Studio - It is a web application for labeling findings from publications, patents, and other free text. It's based on Label Studio, but more functions for labeling findings.
Paper Downloader - A tool for downloading the papers from the pubmed.
Knowledge Graph Editor - This extension is designed for synchronizing findings/knowledges into a biomedgps instance. The findings/knowledges are annotated from literature and other sources using Label Studio.
BioMedGPS - A knowledge graph system with graph neural network for drug discovery, disease mechanism and biomarker screening.
NER & RE Models - A set of models for biomedical named entity recognition and relation extraction.
Text2Knowledges - Extract entities and relationships from free text.

Launch the prophet studio

Advantage: You can make the paper-downloader as a service and it will listen the events of the bucket. When you upload a file to the bucket, the paper-downloader will download the file automatically.

git clone https://github.com/yjcyxky/prophet-studio.git

# Before you launch the prophet studio, you need to modify the `.env.example` file according to your situation, such as <MINIO_ACCESS_KEY>, <MINIO_SECRET_KEY>, <DINGTALK_ACCESS_KEY> etc. and rename it to `.env` file.

cd prophet-studio
docker compose up -d

After that, you can access the label studio by https://<your-ip>:8080 with the related access key and secret key.

FAQs

If you create a new organization, you need to relaunch the prophet studio. Otherwise, the paper-downloader will not work.

cd prophet-studio
docker compose down && docker compose up -d

[Advanced] Setup the label studio

More details on the Label Studio, please refer to the customized label studio, which is a labeling system for biomedical paper.

Create a service account on the miniocloud

Login to the miniocloud (https://:9000)
Click on the Service Accounts tab
Click on Create Service Account
Fill in the form and click on Create [Don't enable Restrict with policy]

Enable https on the miniocloud

Setup the label studio

Create a project on the label studio
Go to the Settings tab
Click on Cloud Storage tab
Click on Add Source Storage and fill in the form

NOTE:

You must use the https url of the miniocloud

You must enable the Use pre-signed URLs option, so the label studio can generate the pre-signed url for the html files automatically.

Prepare the data for the label studio

Each task in the label studio is a single HTML file
The HTML files must be stored in the miniocloud
You must create a tasks.json file that contains the list of the HTML files
Each task in the tasks.json file must have the following format:

# `html` is a variable that contains the path to the html file, you need to keep it same with your labeling interface settings.
# Label studio will automatically fetch the html file from the miniocloud
[
    {
        "html": "s3:https://<bucket_name>/<path_to_the_html_file>",
        "pdf": "<embed src='https://<bucket_name>.minio.<region>.miniocloud.com/<path_to_the_pdf_file>' width='100%' height='100%' type='application/pdf'>"
    }
]

# NOTE:
# If you want to show the pdf file on Chrome, it may complain about the sandboxing. I've no idea how to fix it. But it works on Firefox.

How to prepare your data

To choose a query string that can be used to query pubmed for the papers you want to label
To Prepare a yaml file which contains two fields: query_str, author and download_pdf. You need to use plain text editor or code editor to create the yaml file. The yaml file must have the following format:
```
query_str: "your query string"
author: "your name"
download_pdf: true
```
NOTE:
1. The query_str field is used to query pubmed for the papers you want to label
2. The author field is used to identify who is labeling the papers
3. The download_pdf field is used to indicate whether the pipeline should download the pdf files of the papers automatically. If you set it to true, the pipeline will download the pdf files of the papers automatically. If you set it to false, the pipeline will not download the pdf files of the papers automatically. You can download the pdf files manually by paper-downloader tool. After downloading all pdfs of the papers, you need to reimport the metadata file to the label studio. The metadata file will be stored in the metadata/<project_name> folder. The pdf files will be stored in the pdf folder. [NOTE: Not all the papers have pdf files, if you see a paper that doesn't have a pdf file on label studio, it means the pipeline couldn't download the pdf file of the paper automatically. You can download the pdf file manually and upload it to the pdf folder]
4. The download_pdf field is optional. If you don't set it, the pipeline will not download the pdf files of the papers automatically.
To upload the yaml file to the config/<project_name> folder
Wait a few minutes for the pipeline to fetch the papers and generate the metadata file for the papers. The metadata file will be stored in the metadata/<project_name> folder. If you want to check the progress of the pipeline, you can check the dingtalk group. But if you enable the download_pdf field, the pipeline will take a long time to download the pdf files of the papers. So you can check the progress of the pipeline by checking the pdf/<project_name> folder. The pipeline will download the pdf files of the papers one by one. If you see a pdf file in the pdf/<project_name> folder, it means the pipeline has finished downloading the pdf file of the paper.
If you see a notification in the dingtalk group, it means the pipeline has finished fetching the metadata of papers. After that, the system administrator will also get the notification and upload the metadata file to the label studio and download all the pdf files of the papers. The pdf files will be stored in the pdf/<project_name> folder. [NOTE: Not all the papers have pdf files, if you see a paper that doesn't have a pdf file on label studio, it means the pipeline couldn't download the pdf file of the paper automatically. You can download the pdf file manually and upload it to the pdf folder]
To start labeling the papers;

Templates

For Abstract Labeling

<View>
  <View>
    <Style>
        .rating div { margin-top: 0px !important; margin-bottom: 0px !important; }
        .rating h4 { font-size: 14px; font-weight: 400; }
        .htx-richtext { text-align: justify; }
        .text-area { width: 100%; }
        .title { display: inline-flex; margin-right: 10px; }
        .lsf-labels { margin: 0px !important; }
        .lsf-main-view { background-color: #ffffff; }
        .lsf-richtext__iframe { max-height: 400px; }
    </Style>
    <View style="display:flex;align-items:start;flex-direction:column;">
      <View style="margin-bottom: 5px;">
        <Filter toName="label" minlength="0" name="filter"/>
      </View>
      <Labels name="label" toName="text" showInline="true">
        <Label value="gene" background="red"/>
        <Label value="protein" background="orange"/>
        <Label value="disease" background="olive"/>
        <Label value="symptom" background="indigo"/>
        <Label value="metabolite" background="green"/>
        <Label value="pathway" background="violet"/>        
        <Label value="anatomy" background="grey"/>
        <label value="microbe" background="yellow"/>
        <label value="chemical" background="black"/>
        <label value="biological_process" background="purple"/>
        <label value="cellular_component" background="pink"/>
        <label value="molecular_function" background="olive"/>
        <Label value="key_sentence" background="blue"/>
      </Labels>
      <View className="title">
        <Header value="$title" underline="true"/>
      </View>
      <View className="rating" style="display: flex; flex-direction: row; align-items: center;">
        <Header value="$journal | IF: $impact_factor | PMID: $pmid" underline="false" />
        <Choices name="valid" showInline="true" toName="text" choice="single-radio">
          <Choice alias="YES" value="YES"/>
          <Choice alias="NO" value="NO"/>
          <Choice alias="UNKNOWN" value="Unknown"/>
        </Choices>
        <Rating name="rating" toName="text" maxRating="5" icon="star" size="medium"/>
      </View>
      <View style="width:100%;">
        <Text name="text" valueType="text" value="$abstract"/>
      </View>
      <View className="text-area">
        <Choices name="full_text_required" showInline="true" toName="text" choice="single-radio">
          <Choice alias="YES" value="YES (Full Paper Required)" hint="Need to label full paper."/>
          <Choice alias="NO" value="NO" selected="true" />
        </Choices>
        <TextArea name="notes" toName="text" required="false" maxSubmissions="1" rows="5" placeholder="Input your notes."/>
      </View>
      <View>
        <Relations>
          <Relation value="biomarker"/>
          <Relation value="treats"/>
          <Relation value="associated_with"/>
          <Relation value="upregulated_in"/>
          <Relation value="downregulated_in"/>
          <Relation value="activated_by"/>
          <Relation value="inhibited_by"/>
          <Relation value="reduced_by"/>
          <Relation value="increased_by"/>
          <Relation value="alleviated_by"/>
          <Relation value="induced_by"/>
        </Relations>
      </View>
    </View>
  </View>
</View>

For Full Text Labeling

<View>
  <View>
    <Style>
        .rating div { margin-top: 0px !important; margin-bottom: 0px !important; }
        .htx-richtext { text-align: justify; }
        .text-area { width: 100%; }
        .title { display: inline-flex; margin-right: 10px; }
        .lsf-labels { margin: 0px !important; }
        .lsf-main-view { background-color: #ffffff; }
        .lsf-richtext__iframe { max-height: 400px !important; }
        .html .lsf-richtext__iframe { min-height: 600px !important; }
    </Style>
    <View style="display:flex;align-items:start;flex-direction:column;">
      <View style="margin-bottom: 5px;">
        <Filter toName="label" minlength="0" name="filter"/>
      </View>
      <Labels name="label" toName="html" showInline="true">
        <Label value="gene" background="red"/>
        <Label value="protein" background="orange"/>
        <Label value="disease" background="olive"/>
        <Label value="metabolite" background="green"/>
        <Label value="pathway" background="violet"/>        
        <Label value="anatomy" background="grey"/>
        <label value="microbe" background="yellow"/>
        <label value="chemical" background="black"/>
        <label value="biological_process" background="purple"/>
        <label value="cellular_component" background="pink"/>
        <label value="molecular_function" background="olive"/>
        <Label value="key_sentence" background="blue"/>
      </Labels>
      <View className="title">
        <Header value="$title" underline="true"/>
      </View>
      <View className="rating" style="display: flex; flex-direction: row; align-items: center;">
        <Header value="$journal | IF: $impact_factor | PMID: $pmid" underline="false" />
        <Choices name="valid" showInline="true" toName="html" choice="single-radio">
          <Choice alias="YES" value="YES"/>
          <Choice alias="NO" value="NO"/>
          <Choice alias="UNKNOWN" value="Unknown"/>
        </Choices>
        <Rating name="rating" toName="html" maxRating="5" icon="star" size="medium"/>
      </View>
      <View style="width:100%;" className="html">
       	<HyperText name="html" value="$html" valueType="url" />
      </View>
      <View className="text-area">
        <TextArea name="notes" toName="html" required="false" maxSubmissions="1" rows="5" placeholder="Input your notes."/>
      </View>
      <View>
        <Relations>
          <Relation value="associated_with"/>
          <Relation value="upregulated_in"/>
          <Relation value="downregulated_in"/>
          <Relation value="activated_by"/>
          <Relation value="inhibited_by"/>
          <Relation value="reduced_by"/>
          <Relation value="increased_by"/>
          <Relation value="alleviated_by"/>
          <Relation value="induced_by"/>
        </Relations>
      </View>
    </View>
  </View>
</View>

Quickstart with an ML backend

Launch the ML backend

Label Studio supports integration with machine learning backends. You can use a machine learning backend to pre-label your data, or to automatically label your data as it is uploaded to Label Studio. For more information, see the Label Studio documentation.

To start an example machine learning backend with Docker Compose, do the following:

Make sure port 9090 is available (or override it with –port parameter).
Clone the current git repository. From the command line, run the following:

git clone https://github.com/heartexlabs/label-studio-ml-backend

Change to the directory with the Docker Compose configuration file. From the command line, run the following: bash

cd models/ner

Start Docker Compose. From the command line, run the following:

docker-compose up

The machine learning backend server starts becomes available at https://localhost:9090. You can now add the machine learning backend to Label Studio and set up your project.

NOTICES:

Current model only support one to_name field in the configuration of the label studio. Otherwise, it will complain about the assert len(self.parsed_label_config) == 1 error. One solution is that you define one to_name field first and then add other to_name fields after the model added to the label studio.

The model will only label the text field. Don't use HyperText field in the configuration of the label studio. Otherwise, it will complain about the Cannot convert string to int error.

Add an ML backend using the Label Studio UI

In the Label Studio UI, open the project that you want to use with your ML backend.

Click Settings > Machine Learning.
Click Add Model.
Type a Title for the model and provide the URL for the ML backend. For example, https://localhost:9090.

NOTE: If you run the ML backend and Label Studio in docker containers or on different machines, you must use the IP address or url of the machine that runs the ML backend. For example, https://example.com:9090.
- (Optional) Type a description.
- (Optional) Select Use for interactive preannotation. [NOTE: Don't use this option, if your model is not accurate enough, it will mess up your data.]
Click Validate and Save.

If you see any errors, see Troubleshoot machine learning.

Name		Name	Last commit message	Last commit date
Latest commit History 30 Commits
assets		assets
conf		conf
data		data
datasets		datasets
docs		docs
example		example
models		models
prophet_studio		prophet_studio
.env.example		.env.example
.gitignore		.gitignore
.gitmodules		.gitmodules
Makefile		Makefile
README.md		README.md
docker-compose.yml		docker-compose.yml

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Prophet Studio

Why Prophet Studio?

Components

Launch the prophet studio

FAQs

[Advanced] Setup the label studio

Create a service account on the miniocloud

Enable https on the miniocloud

Setup the label studio

Prepare the data for the label studio

How to prepare your data

Templates

For Abstract Labeling

For Full Text Labeling

Quickstart with an ML backend

Launch the ML backend

Add an ML backend using the Label Studio UI

About

Releases

Packages

Contributors 2

Languages

open-prophetdb/prophet-studio

Folders and files

Latest commit

History

Repository files navigation

Prophet Studio

Why Prophet Studio?

Components

Launch the prophet studio

FAQs

[Advanced] Setup the label studio

Create a service account on the miniocloud

Enable https on the miniocloud

Setup the label studio

Prepare the data for the label studio

How to prepare your data

Templates

For Abstract Labeling

For Full Text Labeling

Quickstart with an ML backend

Launch the ML backend

Add an ML backend using the Label Studio UI

About

Topics

Resources

Stars

Watchers

Forks

Releases

Packages 0

Contributors 2

Languages

Packages