Biomedical Entity Recognition

By for September 18, 2017

Report Abuse
Biomedical named entity recognition is a critical step for complex biomedical NLP tasks such as: * Extraction of drug, disease, symptom mentions from electronic health records (EHR) and medical articles. * Drug discovery * Understanding the interactions between different entity types such as drug-drug interaction, drug-disease relationship and gene-protein relationship. Our use case scenario focuses on how a large amount of unstructured unlabeled data corpus such as PubMed article abstracts can be analyzed to train a domain-specific word embedding model. Then the output embeddings are considered as automatically generated features to train a neural entity extraction model using Keras with TensorFlow deep learning framework as backend and a small amount of labeled data.
The aim of this real-world scenario is to highlight how to use Azure Machine Learning Workbench to solve a complicated NLP task such as entity extraction from unstructured text. Here are the key points addressed: 1. How to train a neural word embeddings model on a text corpus of about 18 million PubMed abstracts using [Spark Word2Vec implementation]( 2. How to build a deep Long Short-Term Memory (LSTM) recurrent neural network model for entity extraction on a GPU-enabled Azure Data Science Virtual Machine (GPU DSVM) on Azure. 2. Demonstrate that domain-specific word embeddings model can outperform generic word embeddings models in the entity recognition task. 3. Demonstrate how to train and operationalize deep learning models using Azure Machine Learning Workbench. The following capabilities within Azure Machine Learning Workbench: * Instantiation of [Team Data Science Process (TDSP) structure and templates]( * Automated management of your project dependencies including the download and the installation * Execution of code in Jupyter notebooks as well as Python scripts. * Run history tracking for Python files. * Execution of jobs on remote Spark compute context using HDInsight Spark 2.1 clusters. * Execution of jobs in remote GPU VMs on Azure. * Easy operationalization of deep learning models as web-services hosted on Azure Container Services. The **detailed documentation** for this scenario including the step-by-step walk-through: For code samples, click the **View Project** icon on the right and visit the project GitHub repository. ## Key components needed to run this example: * An Azure [subscription]( * Azure Machine Learning Workbench with a workspace created. See [installation guide]( * To run this scenario with Spark cluster, provision [Azure HDInsight Spark cluster]( (Spark 2.1 on Linux (HDI 3.6)) for scale-out computation. To process the full amount of MEDLINE abstracts discussed below, We recommend having a cluster with: * a head node of type [D13_V2]( * at least four worker nodes of type [D12_V2]( * To maximize performance of the cluster, we recommend to change the parameters spark.executor.instances, spark.executor.cores, and spark.executor.memory by following the instructions [here]( and editing the definitions in "custom spark defaults" section. * You can run the entity extraction model training locally on a [Data Science Virtual Machine (DSVM)]( or in a remote Docker container in a remote DSVM. * To provision DSVM for Linux (Ubuntu), follow the instructions [here]( We recommend using [NC6 Standard (56 GB, K80 NVIDIA Tesla)](