Deep Learning on High-Throughput Transcriptomics to Predict Drug-Induced Liver Injury

Area of Science:

• Pharmacogenomics
• Computational Biology
• Toxicology

Background:

• Drug-induced liver injury (DILI) is a major cause of drug attrition and market withdrawal.
• High-throughput transcriptomic data and deep learning offer potential for improved DILI prediction.

Purpose of the Study:

• To develop and validate a deep neural network (DNN) model for predicting DILI using transcriptomic profiles.
• To assess the model's performance against conventional machine learning algorithms and its utility for oncology drugs.

Main Methods:

• An eight-layer DNN model was developed using the LINCS L1000 dataset and DILIst annotations.
• Model performance was evaluated using Monte Carlo cross-validation, permutation testing, and an independent validation set.
• Comparison with K-nearest neighbors (KNN), Support Vector Machine (SVM), and Random Forest (RF) algorithms.

Main Results:

• The DNN model achieved an AUC of 0.802 (training) and 0.798 (independent validation), outperforming other ML algorithms.
• Balanced accuracy was 0.741 (training) and 0.721 (independent validation), with sensitivity 0.839 and specificity 0.603.
• The DNN model demonstrated superior predictive performance for oncology drugs and identified genes relevant to DILI mechanisms.

Conclusions:

• The developed DNN model is a promising tool for early DILI prediction in pre-clinical settings.
• The model's ability to predict DILI, particularly for oncology drugs, can aid in drug development.
• Functional analysis of predictive genes provides insights into DILI pathogenesis.