AI-Assisted Analysis of Clinical Trials to Identify Failed Drugs for Revival via Drug Delivery Technologies

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Abstract Numerous drug candidates showed great promise in experimental and preclinical investigations but failed in clinical trials. One cause to this problem may be nonspecific drug biodistribution, leading to adverse effects in healthy organs. Drug delivery technologies are developed to overcome this exact problem by navigating drugs more selectively to the disease sites. Therefore, they hold the promise to revive at least some failed drugs. However, it is challenging to identify such candidates revivable by drug delivery: first, there lacks systemic survey and documentation of drug biodistribution in patients during the trials; second, clinical trial results, especially the failed ones, are often reported in unstructured and/or inconsistent language. Here, we harnessed the recent advances in artificial intelligence (AI) agents and large language models (LLMs) to tackle with the second problem. An OpenAI-based framework was developed to systemically analyze terminated trials on ClinicalTrials.gov. This framework integrated structured extraction of clinical trial information, disease-site inference with evaluation based on predefined screening criteria, which is unwanted biodistribution (UB failure) reflected by the difference in the anatomical locations between adverse effects and targeted diseases. We also paired this with the analysis of drug types and physicochemical properties (e.g. molecular weight and hydrophobicity). Using a validation set, the model achieved accuracies of 89.69% for UB failure classification, 91.75% for small molecule drug classification, and 88.66% for hydrophobicity classification. We then applied this analysis to all terminated ClinicalTrials.gov trials, which identified 817 trials possibly failed due to UB, and top ten small molecule drugs likely suitable for our peptide-guided delivery platform. Overall, this study may provide a new direction to apply AI technologies for drug development, and inspire future investigations of clinical reports and public information from the perspective of delivery technology applications.