Integrating Drug Target Information in Deep Learning Models to Predict the Risk of Adverse Events in Patients with...

Area of Science:

• Computational psychiatry and pharmacology
• Biomedical informatics and machine learning

Background:

• Comorbid post-traumatic stress disorder (PTSD) and alcohol use disorder (AUD) significantly increase the risk of adverse health outcomes.
• Limited effective treatment options are available for individuals with co-occurring PTSD and AUD.
• Predicting and preventing adverse events in this high-risk population remains a critical challenge.

Purpose of the Study:

• To develop a novel deep learning model, T-DeepBiomarker, for predicting adverse outcomes in PTSD + AUD patients.
• To integrate multimodal data, including drug targets and social determinants of health (SDoH), for enhanced prediction.
• To identify potential therapeutic medications for mitigating adverse events in this patient group.

Main Methods:

• Utilized electronic medical record (EMR) data from 5565 PTSD + AUD patients.
• Developed T-DeepBiomarker by integrating lab results, drug targets, comorbidities, and SDoH data.
• Trained the model to predict adverse events (opioid use disorder, suicidal behaviors, depression, death) within three months.

Main Results:

• T-DeepBiomarker achieved high predictive performance with an AUROC of 0.94 for adverse outcomes.
• Identified several candidate medications (e.g., Acamprosate, Semaglutide) targeting significant proteins.
• These medications show potential for reducing the risk of adverse events in PTSD + AUD patients.

Conclusions:

• T-DeepBiomarker accurately predicts adverse outcomes in patients with co-occurring PTSD and AUD.
• The study identifies promising candidate drugs for pharmacotherapy in this high-risk population.
• Findings warrant further investigation to advance treatment strategies for PTSD + AUD.