DrugResponse2Vec

A Pytorch Implementation of paper

DrugResponse2vec: Representation of Drug in Vector Space based on Drug-Cell line Response
Jeon and Park et al., 2018

Abstract

As more data available, it has become possible to data-driven new drug discovery pipeline. We, therefore, propose a model for predicting a similarity score between two drugs based on the differential gene expression patterns of the two drugs. Our model trains a Siamese neural network that takes fingerprints of a pair of two drugs and predicts its similarity score.

Pipeline

Project Structure

./main.py : Run experiments with arguments.
./tasks/drug_run.py : Functions for running experiments.
./tasks/drug_task.py : Preprocess datasets.
./models/drug_model.py : LSTM Siamese network.

Run Code

Requirement

Python 3.4.3
Pytorch 0.3.0

Preprocess dataset

Before preprocessing, you should create a directory ./tasks/data/drug/ and unzip files downloaded from here.

# This will create a pickle file ./data/drug/drug(tmp).
$ cd tasks/
$ python3 drug_task.py

Train/test LSTM Siamese network.

usage: main.py [-h] [--data-path DATA_PATH] [--drug-dir DRUG_DIR]
               [--drug-files DRUG_FILES] [--pair-dir PAIR_DIR]
               [--checkpoint-dir CHECKPOINT_DIR] [--model-name MODEL_NAME]
               [--print-step PRINT_STEP] [--validation-step VALIDATION_STEP]
               [--train TRAIN] [--pretrain PRETRAIN] [--valid VALID]
               [--test TEST] [--resume RESUME] [--debug DEBUG]
               [--save-embed SAVE_EMBED] [--save-prediction SAVE_PREDICTION]
               [--save-pair-score SAVE_PAIR_SCORE] [--top-only TOP_ONLY]
               [--embed-d EMBED_D] [--batch-size BATCH_SIZE] [--epoch EPOCH]
               [--learning-rate LEARNING_RATE] [--weight-decay WEIGHT_DECAY]
               [--grad-max-norm GRAD_MAX_NORM] [--grad-clip GRAD_CLIP]
               [--binary BINARY] [--hidden-dim HIDDEN_DIM]
               [--drug-embed-dim DRUG_EMBED_DIM] [--lstm-layer LSTM_LAYER]
               [--lstm-dr LSTM_DR] [--char-dr CHAR_DR] [--bi-lstm BI_LSTM]
               [--linear-dr LINEAR_DR] [--char-embed-dim CHAR_EMBED_DIM]
               [--s-idx S_IDX] [--rep-idx REP_IDX] [--dist-fn DIST_FN]
               [--seed SEED] [--g_layer G_LAYER] [--g_hidden_dim G_HIDDEN_DIM]
               [--g_out_dim G_OUT_DIM] [--g_dropout G_DROPOUT]

# Train/test Siamese network on Fingerprint dataset (regression)
$ python3 main.py --rep-idx 2 --train True --test True

Experimental Results

CMap Score Regression

DrugResponse2vec

Testset Split	Overall CorrCoef	Top/Bottom 1% CorrCoef	AUROC
Total	0.447	0.877	0.693
KK	0.606	0.932	0.777
KU	0.340	0.737	0.639
UU	0.120	0.240	0.555

CMap Score Binary Classification

DrugResponse2vec

Testset Split	Precision	Recall	F1 Score
Total	0.686	0.780	0.730
KK	0.722	0.880	0.793
KU	0.661	0.720	0.689
UU	0.646	0.658	0.652

Liscense

Apache License 2.0

miyoungko / drug-siamese Goto Github PK

drug-siamese's Introduction

DrugResponse2Vec

Abstract

Pipeline

Project Structure

Run Code

Requirement

Preprocess dataset

Train/test LSTM Siamese network.

Experimental Results

CMap Score Regression

DrugResponse2vec

CMap Score Binary Classification

DrugResponse2vec

Liscense

drug-siamese's People

Contributors

Forkers

Recommend Projects

Recommend Topics

Recommend Org