Work

LLM-assisted large-scale curation of peptide clinical trials maps trends in therapeutic development and trial outcomes

Year: 2026

Type: preprint

Abstract: Interest in peptide-based therapeutics has rapidly grown in recent decades, with peptide drugs promising high target specificity and low immunogenicity. However, structural concerns, such as in vivo instability and poor membrane permeability can bottleneck clinical translation, and large-scale insights into peptide drug trial outcomes and design features remain limited. Here, we present and validate an automated, large language model-assisted data curation workflow that demonstrates annotation agreement (AC1 scores) across clinical trial records comparable to volunteer-based annotations, including delivery mode (0.75), peptide sequence (0.70), trial outcome (0.60), and reason-for-failure (0.90) labels. Using this framework, we compile and analyse a comprehensive historical dataset of 6,834 peptide clinical trials. We show that injectable delivery modalities predominate among peptide therapeutics (73.47%), that peptide pharmaceutical trials have phase-wise success rates of 26.36% for Phase 1 trials, 33.40% for Phase 2, and 57.74% for Phase 3, and that peptide drugs fail due to toxicity or safety concerns far less often than a general drug benchmarking set in late-stage trials (4.38% vs. 17.15% for Phase 3 trials). These findings clarify key patterns and challenges in peptide translation and drug development and offer a scalable methodology for large-scale biomedical data curation.

Source: Research Square

DOI: 10.21203/rs.3.rs-10618534/v1