Drug-Drug Interaction

.templates/config.json

@@ -13,6 +13,13 @@
         "evalOptions": "--eval CategoricalEvaluator -D categoricalevaluator.categoryindexes=1 -D eval.includeobs=true",
         "weightLearningOptions": ""
     },
+    "drug-drug-interaction": {
+        "dataURL": "https://linqs-data.soe.ucsc.edu/public/psl-examples-data/drug-drug-interaction.zip",
+        "weightLearning": true,
+        "pslOptions": "--int-ids --eval AUCEvaluator -D aucevaluator.threshold=0.4",
+        "evalOptions": "--eval DiscreteEvaluator -D discreteevaluator.threshold=0.4",
+        "weightLearningOptions": "GuidedRandomGridSearch -D gridsearch.weights=0.01:0.1:1.0 -D randomgridsearch.maxlocations=75 -D weightlearning.evaluator=AUCEvaluator"
+    },
     "entity-resolution": {
         "dataURL": "https://linqs-data.soe.ucsc.edu/public/psl-examples-data/entity-resolution/entity-resolution-small.zip",
         "weightLearning": true,

drug-drug-interaction/README.md

@@ -0,0 +1,39 @@
+### Drug-Drug Interaction
+
+## Problem
+In this example, we attempt to infer unknown drug drug interactions from a network of multiple drug-based similarities and known interactions.
+
+## Dataset
+There are three datasets which are derived from [DrugBank 4.3](https://go.drugbank.com/downloads/archived) and [drugs.com](Drugs.com). Each dataset below contain seven drug–drug similarities. Four of these similarity measures are drug-based: Chemical-based, Ligand-based, Side-effect-based and Annotation-based. Three similarities are between drug targets and computed by aggregating over known targets for the drugs: Sequence-based, PPI network-based, and Gene Ontology-based.
+- general-interactions: This dataset has a total of 4,293 interactions across 315 drugs. There is also a file DrugBankIDs, which maps integer IDs to DrugIDs. Each fold share the same similarity data.
+- crd-interactions: This dataset has a total of 10,106 CRD interactions across 807 drugs. These IDs are anonymized (so there is no DrugBankID mapping). Each fold uses different similarity data (since blocking is stricter in the crd setting).
+- ncrd-interactions: This dataset has a total of 45,737 NCRD interactions across 807 drugs. These IDs are also anonymized. Each fold uses same similarity data.
+
+## Experimental Setup
+The default settings for the run script is for the dataset "general-interactions". Therefore, the evaluator thresholds must be changed when running other datasets. For example, `./run.sh -D discreteevaluator.threshold=<NEW_THRESHOLD> aucevaluator.threshold=<NEW_THRESHOLD>` will use the new values NEW\_THRESHOLD.
+
+## Weight Learning
+For each dataset, we used a Guided Random Grid Search to tune the weights. This produces slightly better results than the original weight learning method. To produce the same results from the original experiment, then disable weight learning and use the default hardcoded weights.
+
+## Origin
+This example is based on the work from [A Probabilistic Approach for Collective Similarity-Based Drug-Drug Interaction Prediction](https://linqs.org/publications/#id:sridhar-bio16). This [repo](https://bitbucket.org/linqs/psl-drug-interaction-prediction/src/master/) contains the original data and experiments from the paper. This example contains data which have been preprocessed and dumped from the original experiment. To reference the original work, please use this citation:
+
+```
+@article{sridhar:bio16,
+    title = {A Probabilistic Approach for Collective Similarity-Based Drug-Drug Interaction Prediction},
+    author = {Dhanya Sridhar and Shobeir Fakhraei and Lise Getoor},
+    journal = {Bioinformatics},
+    year = {2016},
+    publisher = {Oxford},
+    pages = {3175--3182},
+    volume = {32},
+    number = {20},
+}
+```
+
+## Keywords
+ - `cli`
+ - `evaluation`
+ - `inference`
+ - `real data`
+ - `weight learning`

drug-drug-interaction/cli/drug-drug-interaction-eval.data

@@ -0,0 +1,27 @@
+predicates:
+    ATCSimilarity/2: closed
+    SideEffectSimilarity/2: closed
+    GOSimilarity/2: closed
+    ligandSimilarity/2: closed
+    chemicalSimilarity/2: closed
+    seqSimilarity/2: closed
+    distSimilarity/2: closed
+    validInteraction/2: closed
+    interacts/2: open
+
+observations:
+   ATCsimilarity: ../data/drug-drug-interaction/general-interactions/00/eval/similarity_atc.txt
+   chemicalSimilarity: ../data/drug-drug-interaction/general-interactions/00/eval/similarity_chemical.txt
+   distSimilarity: ../data/drug-drug-interaction/general-interactions/00/eval/similarity_dist.txt
+   GOSimilarity: ../data/drug-drug-interaction/general-interactions/00/eval/similarity_go.txt
+   ligandSimilarity: ../data/drug-drug-interaction/general-interactions/00/eval/similarity_ligand.txt
+   seqSimilarity: ../data/drug-drug-interaction/general-interactions/00/eval/similarity_seq.txt
+   SideEffectSimilarity: ../data/drug-drug-interaction/general-interactions/00/eval/similarity_sideeffects.txt
+   validInteraction: ../data/drug-drug-interaction/general-interactions/00/eval/valid_interactions_obs.txt
+   interacts: ../data/drug-drug-interaction/general-interactions/00/eval/interacts_obs.txt
+
+targets:
+   interacts: ../data/drug-drug-interaction/general-interactions/00/eval/interacts_target.txt
+
+truth:
+   interacts: ../data/drug-drug-interaction/general-interactions/00/eval/interacts_truth.txt

drug-drug-interaction/cli/drug-drug-interaction-learn.data

@@ -0,0 +1,27 @@
+predicates:
+    ATCSimilarity/2: closed
+    SideEffectSimilarity/2: closed
+    GOSimilarity/2: closed
+    ligandSimilarity/2: closed
+    chemicalSimilarity/2: closed
+    seqSimilarity/2: closed
+    distSimilarity/2: closed
+    validInteraction/2: closed
+    interacts/2: open
+
+observations:
+   ATCsimilarity: ../data/drug-drug-interaction/general-interactions/00/learn/similarity_atc.txt
+   chemicalSimilarity: ../data/drug-drug-interaction/general-interactions/00/learn/similarity_chemical.txt
+   distSimilarity: ../data/drug-drug-interaction/general-interactions/00/learn/similarity_dist.txt
+   GOSimilarity: ../data/drug-drug-interaction/general-interactions/00/learn/similarity_go.txt
+   ligandSimilarity: ../data/drug-drug-interaction/general-interactions/00/learn/similarity_ligand.txt
+   seqSimilarity: ../data/drug-drug-interaction/general-interactions/00/learn/similarity_seq.txt
+   SideEffectSimilarity: ../data/drug-drug-interaction/general-interactions/00/learn/similarity_sideeffects.txt
+   validInteraction: ../data/drug-drug-interaction/general-interactions/00/learn/valid_interactions_obs.txt
+   interacts: ../data/drug-drug-interaction/general-interactions/00/learn/interacts_obs.txt
+
+targets:
+   interacts: ../data/drug-drug-interaction/general-interactions/00/learn/interacts_target.txt
+
+truth:
+   interacts: ../data/drug-drug-interaction/general-interactions/00/learn/interacts_truth.txt

drug-drug-interaction/cli/drug-drug-interaction.psl

@@ -0,0 +1,12 @@
+5: ATCSimilarity(D1, D2) & interacts(D1, D3) & validInteraction(D1, D3) & validInteraction(D2, D3) & (D2 != D3) & (D1 != D2) -> interacts(D2, D3) ^2
+5: SideEffectSimilarity(D1, D2) & interacts(D1, D3) & validInteraction(D1, D3) & validInteraction(D2, D3) & (D2 != D3) & (D1 != D2) -> interacts(D2, D3) ^2
+5: GOSimilarity(D1, D2) & interacts(D1, D3) & validInteraction(D1, D3) & validInteraction(D2, D3) & (D2 != D3) & (D1 != D2) -> interacts(D2, D3) ^2
+5: ligandSimilarity(D1, D2) & interacts(D1, D3) & validInteraction(D1, D3) & validInteraction(D2, D3) & (D2 != D3) & (D1 != D2) -> interacts(D2, D3) ^2
+5: chemicalSimilarity(D1, D2) & interacts(D1, D3) & validInteraction(D1, D3) & validInteraction(D2, D3) & (D2 != D3) & (D1 != D2) -> interacts(D2, D3) ^2
+5: seqSimilarity(D1, D2) & interacts(D1, D3) & validInteraction(D1, D3) & validInteraction(D2, D3) & (D2 != D3) & (D1 != D2) -> interacts(D2, D3) ^2
+5: distSimilarity(D1, D2) & interacts(D1, D3) & validInteraction(D1, D3) & validInteraction(D2, D3) & (D2 != D3) & (D1 != D2) -> interacts(D2, D3) ^2
+
+// prior
+5: validInteraction(D1,D2) -> !interacts(D1,D2) ^2
+
+interacts(D1, D2) = interacts(D2, D1) .

drug-drug-interaction/cli/run.sh

@@ -0,0 +1,162 @@
+#!/bin/bash
+
+# Options can also be passed on the command line.
+# These options are passed blindly to the PSL CLI.
+# Ex: ./run.sh -D log4j.threshold=DEBUG
+
+readonly THIS_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
+
+readonly PSL_VERSION='3.0.0-SNAPSHOT'
+readonly JAR_PATH="${THIS_DIR}/psl-cli-${PSL_VERSION}.jar"
+readonly RUN_SCRIPT_VERSION='1.3.6'
+
+readonly BASE_NAME='drug-drug-interaction'
+readonly OUTPUT_DIRECTORY="${THIS_DIR}/inferred-predicates"
+
+readonly ADDITIONAL_PSL_OPTIONS='--int-ids --eval AUCEvaluator -D aucevaluator.threshold=0.4'
+readonly ADDITIONAL_WL_OPTIONS='--learn GuidedRandomGridSearch -D gridsearch.weights=0.01:0.1:1.0 -D randomgridsearch.maxlocations=75 -D weightlearning.evaluator=AUCEvaluator'
+readonly ADDITIONAL_EVAL_OPTIONS='--infer --eval DiscreteEvaluator -D discreteevaluator.threshold=0.4'
+
+function main() {
+    trap exit SIGINT
+
+    bash "${THIS_DIR}/../data/fetchData.sh"
+
+    # Make sure we can run PSL.
+    check_requirements
+    fetch_psl
+
+    # Run PSL.
+    run_weight_learning "$@"
+    run_inference "$@"
+}
+
+function run_weight_learning() {
+    echo "Running PSL Weight Learning."
+
+    java -jar "${JAR_PATH}" \
+        --model "${THIS_DIR}/${BASE_NAME}.psl" \
+        --data "${THIS_DIR}/${BASE_NAME}-learn.data" \
+        ${ADDITIONAL_PSL_OPTIONS} ${ADDITIONAL_WL_OPTIONS} "$@"
+
+    if [[ "$?" -ne 0 ]]; then
+        echo 'ERROR: Failed to run weight learning.'
+        exit 60
+    fi
+}
+
+function run_inference() {
+    echo "Running PSL Inference."
+
+    java -jar "${JAR_PATH}" \
+        --model "${THIS_DIR}/${BASE_NAME}-learned.psl" \
+        --data "${THIS_DIR}/${BASE_NAME}-eval.data" \
+        --output "${OUTPUT_DIRECTORY}" \
+        ${ADDITIONAL_PSL_OPTIONS} ${ADDITIONAL_EVAL_OPTIONS} "$@"
+
+    if [[ "$?" -ne 0 ]]; then
+        echo 'ERROR: Failed to run infernce.'
+        exit 70
+    fi
+}
+
+function check_requirements() {
+    local hasWget
+    local hasCurl
+
+    type wget > /dev/null 2> /dev/null
+    hasWget=$?
+
+    type curl > /dev/null 2> /dev/null
+    hasCurl=$?
+
+    if [[ "${hasWget}" -ne 0 ]] && [[ "${hasCurl}" -ne 0 ]]; then
+        echo 'ERROR: wget or curl required to download the jar.'
+        exit 10
+    fi
+
+    type java > /dev/null 2> /dev/null
+    if [[ "$?" -ne 0 ]]; then
+        echo 'ERROR: java required to run project.'
+        exit 13
+    fi
+}
+
+function get_fetch_command() {
+    type curl > /dev/null 2> /dev/null
+    if [[ "$?" -eq 0 ]]; then
+        echo "curl -o"
+        return
+    fi
+
+    type wget > /dev/null 2> /dev/null
+    if [[ "$?" -eq 0 ]]; then
+        echo "wget -O"
+        return
+    fi
+
+    echo 'ERROR: wget or curl not found.'
+    exit 20
+}
+
+function fetch_file() {
+    local url=$1
+    local path=$2
+
+    local name=$(basename "${path}")
+
+    if [[ -e "${path}" ]]; then
+        echo "${name} file found cached, skipping download."
+        return
+    fi
+
+    echo "Downloading ${name} file located at: '${url}'."
+    `get_fetch_command` "${path}" "${url}"
+    if [[ "$?" -ne 0 ]]; then
+        echo "ERROR: Failed to download ${name}."
+        exit 30
+    fi
+}
+
+# Fetch the jar from a remote or local location and put it in this directory.
+# Non-snapshot builds are fetched from Maven Central.
+# For snapshot builds, the local maven cache ($HOME/.m2) is checked first, and then the snapshot deployment servers.
+# Snapshots are fetched from the local maven repo and other builds are fetched remotely.
+function fetch_psl() {
+    if [[ -e "${JAR_PATH}" ]] ; then
+        echo "Using PSL jar found at ${JAR_PATH}. To fetch a new version, delete this cached jar."
+        return
+    fi
+
+    if [[ "${PSL_VERSION}" =~ .*-SNAPSHOT$ ]]; then
+        local snapshotLocalPath="$HOME/.m2/repository/org/linqs/psl-cli/${PSL_VERSION}/psl-cli-${PSL_VERSION}.jar"
+        if [[ -e "${snapshotLocalPath}" ]] ; then
+            echo "Using local PSL snapshot build."
+            cp "${snapshotLocalPath}" "${JAR_PATH}"
+            return
+        fi
+
+        echo "Using remote PSL snapshot build."
+        local snotshotMetadataURL="https://oss.sonatype.org/content/repositories/snapshots/org/linqs/psl-cli/${PSL_VERSION}/maven-metadata.xml"
+        local metadataFilename='._maven-metadata.xml'
+
+        rm -f "${metadataFilename}"
+        fetch_file "${snotshotMetadataURL}" "${metadataFilename}"
+
+        local snapshotDate=$(grep -m 1 'timestamp' "${metadataFilename}" | sed -E 's/^.*>([0-9]+\.[0-9]+)<.*$/\1/')
+        local snapshotNumber=$(grep -m 1 'buildNumber' "${metadataFilename}" | sed -E 's/^.*>([0-9]+)<.*$/\1/')
+        rm -f "${metadataFilename}"
+
+        local baseVersion=$(echo "${PSL_VERSION}" | sed -E 's/-SNAPSHOT$//')
+        local version="${baseVersion}-${snapshotDate}-${snapshotNumber}"
+
+        local snotshotJarURL="https://oss.sonatype.org/content/repositories/snapshots/org/linqs/psl-cli/${PSL_VERSION}/psl-cli-${version}.jar"
+        fetch_file "${snotshotJarURL}" "${JAR_PATH}"
+    else
+        echo "Using remote PSL build."
+        local remoteJarURL="https://repo1.maven.org/maven2/org/linqs/psl-cli/${PSL_VERSION}/psl-cli-${PSL_VERSION}.jar"
+        fetch_file "${remoteJarURL}" "${JAR_PATH}"
+    fi
+}
+
+[[ "${BASH_SOURCE[0]}" == "${0}" ]] && main "$@"

drug-drug-interaction/data/.gitignore

@@ -0,0 +1,3 @@
+*
+!fetchData.sh
+!.gitignore

drug-drug-interaction/data/fetchData.sh

@@ -0,0 +1,93 @@
+#!/bin/bash
+
+readonly THIS_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
+
+readonly DATA_URL='https://linqs-data.soe.ucsc.edu/public/psl-examples-data/drug-drug-interaction.zip'
+readonly DATA_FILE=$(basename "${DATA_URL}")
+readonly DATA_DIR='drug-drug-interaction'
+readonly SCRIPT_VERSION='1.3.6'
+
+function main() {
+    trap exit SIGINT
+
+    cd "${THIS_DIR}"
+
+    check_requirements
+
+    fetch_file "${DATA_URL}" "${DATA_FILE}"
+    extract_zip "${DATA_FILE}" "${DATA_DIR}"
+}
+
+function check_requirements() {
+    local hasWget
+    local hasCurl
+
+    type wget > /dev/null 2> /dev/null
+    hasWget=$?
+
+    type curl > /dev/null 2> /dev/null
+    hasCurl=$?
+
+    if [[ "${hasWget}" -ne 0 ]] && [[ "${hasCurl}" -ne 0 ]]; then
+        echo 'ERROR: wget or curl required to download the jar.'
+        exit 10
+    fi
+}
+
+function get_fetch_command() {
+    type curl > /dev/null 2> /dev/null
+    if [[ "$?" -eq 0 ]]; then
+        echo "curl -o"
+        return
+    fi
+
+    type wget > /dev/null 2> /dev/null
+    if [[ "$?" -eq 0 ]]; then
+        echo "wget -O"
+        return
+    fi
+
+    echo 'ERROR: wget or curl not found.'
+    exit 20
+}
+
+function fetch_file() {
+    local url=$1
+    local path=$2
+
+    local name=$(basename "${path}")
+
+    if [[ -e "${path}" ]]; then
+        echo "${name} file found cached, skipping download."
+        return
+    fi
+
+    echo "Downloading ${name} file located at: '${url}'."
+    `get_fetch_command` "${path}" "${url}"
+    if [[ "$?" -ne 0 ]]; then
+        echo "ERROR: Failed to download ${name}."
+        exit 30
+    fi
+}
+
+
+function extract_zip() {
+    local path=$1
+    local expectedDir=$2
+
+    local name=$(basename "${path}")
+
+    if [[ -e "${expectedDir}" ]]; then
+        echo "Extracted ${name} zip found cached, skipping extract."
+        return
+    fi
+
+    echo "Extracting the ${name} zip"
+    unzip "${path}"
+    if [[ "$?" -ne 0 ]]; then
+        echo "ERROR: Failed to extract ${name}."
+        exit 40
+    fi
+}
+
+[[ "${BASH_SOURCE[0]}" == "${0}" ]] && main "$@"