TrialEnroll: Predicting Clinical Trial Enrollment Success with Deep & Cross Network and Large Language Models

Congratulations! This paper has been accepted at the 2024 ACM-BCB Conference!

Overview

This repository contains the code for the paper "TrialEnroll: Predicting Clinical Trial Enrollment Success with Deep & Cross Network and Large Language Models." The project leverages advanced machine learning techniques to predict the success of clinical trial enrollments.

Requirements

• Python 3.10
• PyTorch 2.3

Data Sources

• ClinicalTrials.gov
• IQVIA Label Data

Getting Started

1. Download the ClinicalTrial Data

cd data
wget https://clinicaltrials.gov/AllPublicXML.zip

2. Decompress the Data File

unzip AllPublicXML.zip -d trials
find trials/* -name "NCT*.xml" | sort > trials/all_xml.txt

3. Download IQVIA Label Data

Download the IQVIA label data from here and place it in the data/ directory. Rename the file trial_outcomes_v1.csv to IQVIA_trial_outcomes.csv.

Data Preprocessing

1. Preprocess Clinical Trial Data

Navigate to the preprocessing directory and run the following scripts:

cd TrialEnroll/preprocess
python collect_age.py
python collect_location.py
python collect_str.py
python collect_time.py
python save_df.py

2. Preprocess LLM Generated Features

Download the Mistral-7B-Instruct model:

huggingface-cli download --resume-download mistralai/Mistral-7B-Instruct-v0.3 --local-dir 7B-Instruct-v0.3

Set the Mistral path in llm_request_MistralInstruct.py. Then, create the necessary directories and run the preprocessing scripts:

cd TrialEnroll/llm_emb
mkdir -p data_llm/disease/MistralInstruct data_llm/drug/MistralInstruct
python preprocess.py
python llm_request_MistralInstruct.py
python embedding.py

3. Prepare Criteria Embedding

cd TrialEnroll
python protocol_encode.py

Model Training

1. Run DCN Preprocessing

python col_preprocessing.py
python stack_features_dcn.py
python hatten_cross.py

Results

The model achieved a PR AUC of 0.7015.

Citation

If you use this code, please cite our paper:

@article{yue2024trialenroll,
  title={Trialenroll: Predicting clinical trial enrollment success with deep \& cross network and large language models},
  author={Yue, Ling and Xing, Sixue and Chen, Jintai and Fu, Tianfan},
  journal={arXiv preprint arXiv:2407.13115},
  year={2024}
}