Skip to content

Latest commit

 

History

History
642 lines (476 loc) · 28 KB

README.md

File metadata and controls

642 lines (476 loc) · 28 KB

Build status

Open Targets Post-Pipeline ETL process

OpenTargets ETL pipeline to process Pipeline output in order to obtain a new API shaped entities. For the file platformDataBackend.sc

platformDataBackend.sc

Requirements

  1. OpenJDK 8/11
  2. scala 2.12.x (through SDKMAN is simple) Scala Doc
  3. ammonite REPL
  4. Input resources from PIS (src/main/resources/conference.conf)

Notes on Java version

Either Java 8 or 11 can be used to build and run the project, but if you intend to use the compiled jar file on Dataproc you must use Java 8. To avoid this problem altogether, do not use native Java methods unless strictly necessary.

Run the ETL with the Makefile automation

Step 1. Configuration

Configure the run using the profile. You should only need to modify the Ensembl, Chembl and data versions. Other profiles can be created and used by running make set_profile profile=<profile_suffix> where <profile_suffix> would be "dev" if the profile file was called "config.dev".

Step 2. Run

Run make etl_run to set the configs, build the jars, copy them to GCS and run the workflow jar. For PPP releases, simply run make etl_run ppp=true

Configuration

There is a directory called configuration which versions all the release configuration files. To repeat a release use the appropriate configuration file.

Add to your run either commandline or sbt task Intellij IDEA -Dconfig.file=./configuration/<year>/<release>. conf and it will run with appropriate configuration. Missing fields will be resolved with reference.conf.

The most common configuration changes you will need to make are pointing towards the correct input files. To load files we use a structure:

config-field-name {
      format = "csv"
      path = "path to file"
      options = [
        {k: "sep", v: "\\t"}
        {k: "header", v: true}
      ]
    }

The options field configures how Spark will read the input files. Both Json and CSV files have a large number of configurable options, details of which can be found in the documentation

Additional output formats

Typically all outputs should use common.output-format to specify a Spark supported write format ("json", "parquet", etc). Relying on this value allows us to read in pre-computed steps easily.

If you require outputs in additional formats, complete the common.additional-outputs section. Metadata is not generated for these additional sources.

Configuring Spark

If you want to use a local installation of Spark customise the application.conf with the following spark-uri field and adjust any other fields as necessary from the reference.conf template:

spark-uri = "local[*]"
common {
 ...  
}

By default Spark will only write to non-existent directories. This behaviour can be modified using the settings field spark-settings.write-mode using one of the valid inputs "error", "errorifexists", "append", "overwrite", "ignore".

Configuring Loggging

Similarly update the logback configuration. You can add -Dlogback.configurationFile=application.xml and have a logback.xml hanging on your project root or run path. An example log configuration file:

<configuration>

    <appender name="STDOUT" class="ch.qos.logback.core.ConsoleAppender">
        <encoder>
            <pattern>%level %logger{15} - %message%n%xException{10}</pattern>
        </encoder>
    </appender>

    <root level="WARN">
        <appender-ref ref="STDOUT" />
    </root>

    <logger name="io.opentargets.etl" level="DEBUG"/>
    <logger name="org.apache.spark" level="WARN"/>

</configuration>

and try to run one command as follows

java -server -Xms1G -Xmx6G -Xss1M -XX:+CMSClassUnloadingEnabled \
    -Dlogback.configurationFile=application.xml \
    -Dconfig.file=./application.conf \
    -classpath . \
    -jar io-opentargets-etl-backend-assembly-0.1.0.jar [step1 [step2 [...]]]

Create a fat JAR

Simply run one of the following commands:

sbt assembly

or

make build

or for a local build with all the spark dependencies

make build_local

The jar will be generated under target/scala-2.12.10/

Create cluster and launch

Here how to create a cluster using gcloud tool.

The current image version is preview because is the only image that supports Spark3 and Java11.

List of dataproc releases

gcloud beta dataproc clusters create \
    etl-cluster \
    --image-version=2.0-debian10 \
    --properties=yarn:yarn.nodemanager.vmem-check-enabled=false,spark:spark.debug.maxToStringFields=1024,spark:spark.master=yarn \
    --master-machine-type=n1-highmem-16 \
    --master-boot-disk-size=500 \
    --num-secondary-workers=0 \
    --worker-machine-type=n1-standard-16 \
    --num-workers=2 \
    --worker-boot-disk-size=500 \
    --zone=europe-west1-d \
    --project=open-targets-eu-dev \
    --region=europe-west1 \
    --initialization-action-timeout=20m \
    --max-idle=30m

And to submit the job (the jar can also by specified from a gs://...

gcloud dataproc jobs submit spark \
           --cluster=etl-cluster \
           --project=open-targets-eu-dev \
           --region=europe-west1 \
           --async \
           --files=mk-latest.conf \
           --properties=spark.executor.extraJavaOptions=-Dconfig.file=mk-latest.conf,spark.driver.extraJavaOptions=-Dconfig.file=mk-latest.conf \
           --jar=gs://ot-snapshots/etl/jars/io-opentargets-etl-backend-assembly-0.2.5.jar -- disease

where mk-latest.conf is

common {
  output = "gs://ot-snapshots/etl/mk-latest"
}

Using Dataproc workflows

Once PIS has provided all of the necessary inputs, update the configuration files to make use of the new inputs. Create two input files, one using parquet as the output format, and one using json. Consult opentargets-pre-data-releases/22.02.3/ for reference.

If there have been any changes to the ETL create new jars for the literature and ETL project and push them to google storage.

Run the worksheet ./dataproc/workflow-etl.sc which will create a cluster, run all steps and then destroy the cluster.

Step dependencies

Considering only the inputs/outputs of the ETL there are component diagrams available in the 'documentation' directory. etl_current shows the relationships between steps in the ETL. etl_current_full shows those relationships at a more granular level, where inputs and outputs are specifically specified.

By default if a step is run that has dependencies the dependent steps will be run first. For example, if you want to run target, reactome will be run automatically as target depends on reactome.

If you only wish to run a single specified step set the field etl-dag.resolve to false. In this case it is your responsibility to ensure that required inputs are available.

Step notes

Target Validation

Inputs can be provided here where the only logic is to match an ENSG ID against an input column. Any input rows which can't be matched to a target in the platform will be discarded.

Using mouse phenotypes as an example: we match on the column specified in the configuration targetFromSourceId. Any entry which cannot be matched to an ID in the platform's target dataset will be discarded.

Evidence

The Evidence step provides scores for evidence strings by datasource. Detailed information can be found in the documentation. Each input file contains the columns "targetId", "targetFromSourceId", "diseaseId" and "datasourceId", as well as optional extra columns for that data-source.

Scores are calculated using a different formula for each data source configured by the field evidences.data-sources. score-expr (score expression). The score expression is not statically checked for correctness!

Specific data types can be excluded in one of two ways:

  1. Remove the entry from data-sources
  2. Add the value of data-sources.id to the field data-sources-exclude. Using this option allows users to exclude specific source without having to update the reference.conf. For example, to exclude ot_crispr and chembl, update your local configuration with data-sources-exclude = ["ot_crispr", "chembl"].

Output are partitioned by data-sources.id.

Drug

The primary input source of the Drug dataset is ChEMBL. ChEMBL contains almost 2 million molecules, most which are are not 'drugs'. We define a drug to be any molecule that meets one or more of the following criteria:

  • There is at least 1 known indication;
  • There is at least 1 known mechanism of action; or
  • The ChEMBL ID can be mapped to a DrugBank ID.

Inputs

Input Source Notes
chembl-molecule ChEMBL Provided from PIS using ChEMBL private ES server.
chembl-indication ChEMBL Provided from PIS using ChEMBL private ES server.
chembl-mechanism ChEMBL Provided from PIS using ChEMBL private ES server.
chembl-target ChEMBL Provided from PIS using ChEMBL private ES server.
chembl-warning ChEMBL Provided from PIS using ChEMBL private ES server.
disease-etl ETL step 'disease'
target-etl ETL step 'target'
drugbank-to-chembl Drugbank vocabulary File Drugbank Vocabulary

Adding additional resources to enrich data

Addition resources can be specified to enrich the data included in the outputs. The following extension-types are supported:

  • synonyms
  • cross-references

See the sections below for more details on required data structure and limitations.

Additional resources are specified in the configuration as follows:

      drug-extensions = [
        {
          extension-type = <extension type>
          path = <path to file> 
        }
      ]
    }
Synonyms

The Drug Beta step supports the addition of supplementary synonym data sources subject to the following limitations:

  • The input file(s) must be:
    • in json format
    • have a field called 'id' which maps 1-to-1 to either a Drugbank ID or ChEMBL ID. The 'id' field must not contain a mixture of both. If the ID is unknown the data will be discarded silently. If a mixture of ids are provided, it is indeterminate which of the two will be used.
    • have a field called 'synonyms' which are either Strings or arrays of Strings linked to the 'id' field.

The input files are specified in the configuration file under the field drug-extensions. The files can contain additional columns; these will be safely ignored.

New synonyms are added to the 'synonyms' field on the object if they are not already present in either 'synonyms' or 'trade names'. At present it is not possible to add new fields to 'trade names'.

Cross references

The Drug Beta step supports the addition of supplementary cross reference data sources subject to the following limitations:

  • The input file(s) must:
    • in json format
    • have a fields:
      • 'id' which maps 1-to-1 to a ChEMBL ID.
      • 'source'
      • 'reference'

For example:

{"id": ..., "source": ..., "reference": ... }

The input files are specified in the configuration file under the field drug-extensions. The files can contain additional columns; these will be safely ignored.

If the source already exists the new references will be appended to the existing ones, provided that the reference is not already present. If the source does not exist it will be created.

Inputs

Inputs are specified in the reference.conf file and include the following:

Name Source
drug-chembl-molecule ChEMBL - Platform Input Support
drug-chembl-indication ChEMBL - Platform Input Support
drug-chembl-mechanism ChEMBL - Platform Input Support
drug-chembl-target ChEMBL - Platform Input Support
drug-drugbank Release annotation file

The Drug step also relies on several other outputs from the ETL:

Name in Drug Field in configuration file
efo disease
gene target
evidence evidence

Outputs

The Drug step writes three files under the common directory specified in the drug.output.path configuration field:

  • drugs
  • mechanismsOfAction
  • Indications

Each of these outputs includes a field id to allow later linkages between them.

Epmc

This step was incorporated into the ETL to generate the EPMC evidence file used in the Evidence step. This was originally managed by the Open Targets Data Team but moved here to reduce release friction.

Inputs

cooccurences files generated through the processing step of the [Literature pipeline](https://github. com/opentargets/platform-etl-literature/).

Configuration

The flag print-metrics optionally logs statistics about the generated data. Data is cached before metrics are generated, so it is not too deleterious to performance to run this.

Baseline Expression

The primary input sources of the baseline expression dataset are

To run the baseline expression step use the example command under Create a fat JAR with expression as the step name.

EBI Search datasets

The primary input sources for generating the EBI Search datasets are *) Diseases *) Targets *) Evidence *) Association Direct Overalls

The step ebisearch will generate two datasets with targetId, diseaseId, approvedSymbol, name and score.

GO

The 'Go' step generates a small lookup table of gene ontologies.

Inputs

Input Source Notes
go-input PIS Provided by PIS from http://geneontology.org/docs/download-ontology/#go_obo_and_owl

The input is a flat file which does not lend itself to columnar processing so its currently a 'preprocessor' step. If more complicated logic becomes required this should be ported. There is also to option of querying the EBI API but this is quite slow and results in a moderately large dataset which we don't otherwise need.

Target

Generates an index of genes, each uniquely identified by an EnsemblID number. There are approximately 61000 genes available.

Configuration

  • hgnc-orthology-species lists the species to include in Target orthologues. The order of this configuration list is significant as it is used to determine the order in which entries appear in the front-end. Items earlier in the list are more closely related to homo sapiens than items further down the list. This is used to select which species will be included in Target > Homologues. If you want to add a species to this list you must also update Platform Input Support to retrieve that species' gene data.

Inputs

Inputs to the ETL are prepared by Platform Input Support (PIS). PIS does some minimal preprocessing, but it is possible to manually retrieve them and run the step locally. If you would like to run this step locally, retrieve the necessary inputs from one of the Open Targets public input buckets. eg. gs://ot-snapshots/... rather than downloading the files directly from the sources listed here which are included for reference.

Consult the reference.conf file to see how to configure the inputs, most of these require only changing the paths to the data. Options for parsing the inputs should not need to be updated.

  1. HGNC

    • https://storage.googleapis.com/open-targets-data-releases/21.02/input/annotation-files/hgnc_complete_set-2021-02-09.json
  2. Ensembl

    • Use Ensembl human gene JSON file (available from: ftp://ftp.ensembl. org/pub/release-102/json/homo_sapiens/homo_sapiens. json) updating the release as required.
    • It can be useful to convert this file to jsonl format. It can be converted with jq -c . genes[] homo_sapiens. json >> homo_sapiens.jsonl. The file is 4GB, so needs a decent machine (min 32GB RAM) for conversion.
  3. Uniprot

    • uniprot:
      • the Uniprot format in flat txt format instead of xml. This is a flat text file and is provided by PIS. Can be downloaded manually from https://www.uniprot. org/uniprot/?query=reviewed%3Ayes%2BAND%2Borganism%3A9606&compress=yes&format=txt
      • The is a conversion tool to create Scala objects in io.opentargets.etl.preprocess.uniprot
    • uniprot-ssl: Uniprot subcellular annotation file. Available from https://www.uniprot.org/locations/?query=*&format=tab&force=true&columns=id&compress=yes
  4. Gene Ontology

    • Requires files available from EBI:
      • Annotation files for human proteins
      • Annotation files for human RNAs
      • File for eco lookup
      • RNAcentral to Ensembl mapping files
  5. Tep

    • Uses files downloaded for tep key in PIS's config.yaml.
  6. NCBI

    • ncbi: Used for synonyms, data available from: ftp://ftp.ncbi.nlm.nih.gov/gene/DATA/GENE_INFO/Mammalia/Homo_sapiens. gene_info.gz
  7. Human Protein Atlas

    • hpa Used for subcellular locations. Data available from HPA's website
    • hpa-sl-ontology: Additional file provided by data team to map HPA locations to subcellular location ontology.
  8. Project Scores

  9. ChEMBL

    • Target index
  10. Gnomad

    • Used for genetic constraints. Data available from Gnomad website
    • The file is in bgz format, this can be converted to csv with gunzip -c input > output.csv.
  11. Homologs

    • Update the release number as required:
      • ftp://ftp.ensembl.org/pub/release-100/tsv/ensembl-compara/homologies/homo_sapiens/Compara.100.protein_default.homologies.tsv.gz
      • ftp://ftp.ensembl.org/pub/release-100/tsv/ensembl-compara/homologies/homo_sapiens/Compara.100.ncrna_default.homologies.tsv.gz
      • ftp://ftp.ensembl.org/pub/release-100/species_EnsemblVertebrates.txt
    • Files generated by PIS: 104_homology_<species>.tsv where '104' is the Ensembl Release. This is a file of name and gene ids to get the correct name for homology gene ids. There will be one for each species.
    • Both the Compara files should be loaded at the same time in target.input.homology-coding-proteins
  12. Chemical Probes

    • Input file prepared by Data team and made available by PIS.
  13. Reactome

  14. Target safety

    • ToxCast data from data team, using latest from [otar-core](gs://otar001-core/TargetSafety/data_files/ToxCast*. tsv)
    • adverseEvents and safetyRisk data available from otar-core.
      • Note: Prior to the 21.09 release this information was stored in a spreadsheet which had not been updated since 2019. This spreadsheet has been converted into json/parquet files using a utility script. This script should not have to be re-run as the spreadsheet should no longer be used. If the underlying data changes it should be modified in the json/parquet files.
  15. Tractability

  16. GenCode

    • This is used to find the canonical transcript IDs for each gene and their relevant exons. This information would ideally come from Ensemble, but that won't be available until Sept 2022 at the earliest. For additional discussion refer to this issue
    • The file is available to download from: ftp://ftp.ebi.ac. uk/pub/databases/gencode/Gencode_human/release_40/gencode.v40.annotation.gff3.gz, updating the versions as necessary.

OpenFDA FAERS DB

The openFDA drug adverse event API returns data that has been collected from the FDA Adverse Event Reporting System (FAERS), a database that contains information on adverse event and medication error reports submitted to FDA.

Data Processing Summary

  1. OpenFDA "FAERS" data is collected from here (~ 1000 files - May 2020)
  2. Stage 1: Pre-processing of this data (OpenFdaEtl.scala):
    • Filtering:
      • Only reports submitted by health professionals (primarysource.qualification in (1,2,3)).
      • Exclude reports that resulted in death (no entries with seriousnessdeath=1).
      • Only drugs that were considered by the reporter to be the cause of the event (drugcharacterization=1).
      • Remove events (but not the whole report which might have multiple events) that are blacklisted (see Blacklist).
    • Match FDA drug names to Open Targets drug names & then map all these back to their ChEMBL id:
      • Open Targets drug index fields: ‘chembl_id’, ‘synonyms’, ‘pref_name’, ‘trade_names’.
      • openFDA adverse event data fields: ‘drug.medicinalproduct’, ‘drug.openfda.generic_name’, ‘drug.openfda.brand_name’, ‘drug.openfda.substance_name’.
    • Generate table where each row is a unique drug-event pair and count the number of report IDs for each pair, the total number of reports, the total number of reports per drug and the total number of reports per event. Using these calculate the fields required for estimating the significance of each event occuring for each drug, e.g. log-likelihood ratio, (llr) (based on FDA LRT method).
  3. Stage 2: Calculate significance of each event for all drugs based on the FDA LRT method (Monte Carlo simulation) (MonteCarloSampling.scala).

Sample output

{"chembl_id":"CHEMBL1231","event":"cardiac output decreased","meddraCode": ..., "count":1,"llr":8.392140045623442,"critval":4.4247991585588675}
{"chembl_id":"CHEMBL1231","event":"cardiovascular insufficiency","meddraCode": ..., "count":1,"llr":7.699049533524681,"critval":4.4247991585588675}

Notice that the JSON output is actually JSONL. Each line is a single result object.

Configuration

The base configuration is under src/main/resources/reference.conf, openfda section.

CHEMBL Drugs

Refers to the drug output this pipeline's drug step.

OpenFDA FAERS Data

This section refers to the path where the OpenFDA FAERS DB dump can be found, for processing.

Blacklisted Events

Path to the file where to find the list of events that need to be removed from the events in OpenFDA FAERS data, as described above. This list is manually curated.

Meddra (optional)

This pipeline uses an optional subset of data from the Medical Dictionary for Regulatory Activities to link the adverse event terms used by the FDA back to their standardised names.

This data is included in the pipeline output provided by Open Targets, but if you are running the pipeline locally you need to download the most recent Meddra release which is subject to licensing restrictions.

In this section, the path to Meddra data release, if available, is specified. Remove key from configuration file if data is not available. As this is proprietary data it must be provided on an optional basis for users who do not have this dataset.

Montecarlo

Specify the number of permutations and the relevance percentile threshhold.

Sampling

This subsection configures the sampling output from this ETL step.

Outputs

ETL Step outputs.

Unfiltered OpenFDA

This is the OpenFDA FAERS data just before running Montecarlo on it, with or without (depending on whether it was provided) Meddra information.

OpenFDA Processing Results

This is the result data from this ETL step.

Sampling

As the FAERS database is very large (~130GB) it might be useful to extract a stratified sample for analysis and testing purposes.

Caution: Execution of the job will be very slow with sampling enabled because of the large amount of data which needs to be written to disk!

The sampling method will return a subset of the original data, with equal proportions of drugs which likely would have had significant results during MC sampling and those that did not. For instance, if there are 6000 unique ChEMBL entries in the entire dataset, and 500 would have had significant adverse effects, and sampling is set to 0.10, we would expect that the sampled data would have around 600 ChEMBL entries, of which 50 may be significant. As the sampling is random, and the significance of an adverse event depends both on the number of adverse events and drugs in a sample, results are not reproducible. Sampling is provided for basic validation and testing.

The sampled dataset is saved to disk, and can be used as an input for subsequent jobs.

Otar Projects

Annotate diseases with Open Targets Projects metadata. (Internal and Partner Preview Platform)

Requires 3 inputs:

  • Output of disease step
  • 2 Google Sheets exported from internal OT data:
    • a lookup from otar project ids to EFO IDs
    • otar project metadata: name, active status and ID

Development environment notes

Scalafmt Installation

A pre-commit hook to run scalafmt is recommended for this repo though installation of scalafmt is left to developers. The Installation Guide has simple instructions, and the process used for Ubuntu 18.04 was:

cd /tmp/  
curl -Lo coursier https://git.io/coursier-cli &&
    chmod +x coursier &&
    ./coursier --help
sudo ./coursier bootstrap org.scalameta:scalafmt-cli_2.12:2.2.1 \
  -r sonatype:snapshots \
  -o /usr/local/bin/scalafmt --standalone --main org.scalafmt.cli.Cli
scalafmt --version # "scalafmt 2.2.1" at TOW

The pre-commit hook can then be installed using:

cd $REPOS/platform-etl-backend
chmod +x hooks/pre-commit.scalafmt 
ln -s $PWD/hooks/pre-commit.scalafmt .git/hooks/pre-commit

After this, every commit will trigger scalafmt to run and --no-verify can be used to ignore that step if absolutely necessary.

Copyright

Copyright 2014-2024 EMBL - European Bioinformatics Institute, Genentech, GSK, MSD, Pfizer, Sanofi and Wellcome Sanger Institute

This software was developed as part of the Open Targets project. For more information please see: http://www.opentargets.org

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.