MatchEngine

The matchengine matches patient clinical and genomic information to trials.

Built with

• MongoDB - NoSQL document database for data storage.
• nose - Python library for unit testing.

All required python libraries can be installed by running pip install -r requirements.txt

User Guide

Step 1: Set up MongoDB

The matchengine was initially developed using MongoDB version 3.2. For MongoDB installation instructions for Linux, Mac OS X, and Windows please visit their installation page.

Step 2: Load data

Patient data

The matchengine expects patient data to be stored in two separate MongoDB collections:

• clinical: Contains clinical attributes like cancer diagnosis and age (see examples/clinical.example.bson for an example)

MRN	SAMPLE_ID	ONCOTREE_PRIMARY_DIAGNOSIS_NAME	BIRTH_DATE	VITAL_STATUS	GENDER
01	SAMPLE-01	Breast Invasive Ductal Carcinoma	1900-01-01	alive	female

• genomic: Contains all genomic variants sequenced from each patient (see examples/genomic.example.csv for an example)

SAMPLE_ID	TRUE_HUGO_SYMBOL	TRUE_PROTEIN_CHANGE	TRUE_VARIANT_CLASSIFICATION	VARIANT_CATEGORY	CNV_CALL	TRUE_TRANSCRIPT_EXON	WILDTYPE
SAMPLE-01	PIK3CA	p.H1047R	Missense_Mutation	MUTATION		8	false

Clinical and genomic files can be imported to MongoDB using the matchengine in CSV, PKL, and JSON format. MongoDB will store these collections in JSON format and is able to export the files again in BSON, JSON, and CSV format. For more information see mongodump and mongoexport

Trial data

The matchengine expects trial data to also be stored in a separate MongoDB collection. Matching information is stored in a nested structure under the root field name "treatment_list". Trials can be imported to MongoDB using the matchengine in YML or JSON format. In YML format, an example of the trial structure would be:

protocol_no: 00-000
nct_id: NCT000
treatment_list:
  step:
  - arm:
    - arm_code: A
      arm_description: 'Example Arm A'
      arm_internal_id: 1
      arm_suspended: N
      dose_level: []
      match:
        - and:
          - clinical:
              oncotree_primary_diagnosis: Breast
              age_numerical: '>=18'
          - or:
            - genomic:
                hugo_symbol: PIK3CA
                variant_category: Mutation
                protein_change: p.H1047R
            - genomic:
                hugo_symbol: TP53
                variant_category: Mutation

There are several genomic variants that can be curated in this way. Beneath is a map detailing how the trial field names correspond to the patient data field names:

trial field name	genomic field name	example
hugo_symbol	TRUE_HUGO_SYMBOL	ERBB2
protein_change	TRUE_PROTEIN_CHANGE	p.T790M
wildcard_protein_change	TRUE_PROTEIN_CHANGE	p.G719
variant_classification	TRUE_VARIANT_CLASSIFICATION	In_Frame_Del
variant_category	VARIANT_CATEGORY	Mutation
exon	TRUE_TRANSCRIPT_EXON	10
cnv_call	CNV_CALL	Heterozygous deletion
wildtype	WILDTYPE	True or False

trial field name	clinical field name	example
oncotree_diagnosis	ONCOTREE_PRIMARY_DIAGNOSIS_NAME	Breast Invasive Ductal Carcinoma
age_numerical	BIRTH_DATE	1900-01-01

variant_classification options:

• Missense_Mutation
• In_Frame_Del
• Nonsense_Mutation
• Splice_Region
• Frame_Shift_Del
• Splice_Site
• In_Frame_Ins

variant_category options:

• Mutation
• Copy Number Variation
• Structural Variation
• Signature

cnv_call options (for '''variant_category: Copy Number Variation''' only)

• Heterozygous deletion
• Homozygous deletion
• Gain
• High level amplification

Our example

To import example data run:

python matchengine.py load -t examples/trial.example.yml -c examples/clinical.example.csv -g examples/genomic.example.csv --mongo-uri ${your_mongo_uri}

• By default, load inserts the data into a database named matchminer.
• For more information on linking your Mongo URI please see these docs. For default mongo shell configurations this will likely be mongodb://localhost:27017
• Default trial file format is YML. To change this specify --trial-format {yml,json,bson}
• Default clinical file format is CSV. To change this specify --trial-format {csv,pkl,bson}

Step 2: Matching

Once your MongoDB is set up you can perform matching by running:

python matchengine.py match --mongo-uri ${your_mongo_uri}

Default output will be a csv file called "results.csv" in your current working directory. You can specify the outpath path and filename of the results by setting the -o flag.
NOTE: If using -o, please specify output directory and filename. You can change the file format of the output to JSON by setting the --json flag.

Unit testing

The matchengine uses nose for unit testing. To run all tests from the repository's root directory:

nosetests tests

Authors

• Zachary Zwiesler
• Priti Kumari
• James Lindsay