EconPapers    
Economics at your fingertips  
 

Pepti-DrugPhAI: Multimodal AI-evaluator for functional pharmacokinetics and toxicity of peptides

Elizaveta A Bogdanova, Timofei F Lukianov, Artem V Chernukhin and Alexey K Shaytan

PLOS Computational Biology, 2026, vol. 22, issue 11, 1-21

Abstract: Peptide drugs exhibit exceptional selectivity for biological targets and high functional activity making them among the most promising therapeutic agents. However, unsatisfactory pharmacokinetic characteristics remain a critical barrier to their implementation. They often become the primary reason for the failure of promising candidates in preclinical and clinical trials. To address this issue, we present Pepti-DrugPhAI – a comprehensive algorithm for the precise prediction of the pharmacokinetics and toxicity of a wide range of peptide compounds. Unlike many existing solutions, the proposed algorithm is multimodal and can simultaneously process both SMILES representations of chemical structures and amino acid sequences. This ensures efficient processing not only of canonical peptides but also of non-canonical modifications, cyclic structures, and peptide-like small molecules. We also created an original dataset for model training integrating data from public databases and relevant scientific publications. The algorithm simultaneously predicts key parameters including membrane permeability (Caco-2 and PAMPA penetration), distribution and excretion (half-life, clearance, volume of distribution), as well as general and specific toxicity profiles (neurotoxicity, hemolytic activity). A key innovation of the study is the integration of functional pharmacokinetic scoring. The model not only predicts ADMET values but also calculates an adaptive suitability rating for a compound in the context of its specific biological activity. We have conducted a comparative analysis of various data encoding strategies implementing a stack of models ranging from classical machine learning methods to modern transformers and language models in classification modes. Results from testing on independent test sets demonstrate high prediction accuracy and interpretability of features making Pepti-DrugPhAI a powerful tool for more robust selection of promising peptide drug candidates. All data and the code are available in the GitHub repository at https://github.com/EABogdanova/Pepti-DrugPhAI.Author summary: Peptide drugs are highly valued in medicine for their ability to target specific disease mechanisms with minimal side effects. However, their development is frequently hindered by poor pharmacokinetic properties—they are rapidly degraded by enzymes in the body and struggle to cross cell membranes. This causes many promising drug candidates to fail in preclinical and clinical trials. To overcome this, scientists use complex chemical modifications, but predicting how these altered molecules will behave in the body remains a major computational challenge. To address this bottleneck, we developed Pepti-DrugPhAI, a multimodal artificial intelligence platform. Unlike traditional tools that look at either the biological sequence or the chemical structure in isolation, our algorithm analyzes both simultaneously. This allows it to accurately evaluate both natural peptides and those with complex synthetic modifications. Furthermore, rather than simply outputting raw numerical predictions, we introduced an adaptive scoring system that evaluates a molecule’s suitability based on its specific therapeutic purpose, such as acute resuscitation or chronic therapy. By bridging the gap between computational chemistry and systems pharmacology, this tool provides researchers with an intuitive framework to prioritize the best drug candidates early in the development pipeline.

Date: 2026
References: Add references at CitEc
Citations:

Downloads: (external link)
https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1014956 (text/html)
https://journals.plos.org/ploscompbiol/article/fil ... 14956&type=printable (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:plo:pcbi00:1014956

DOI: 10.1371/journal.pcbi.1014956

Access Statistics for this article

More articles in PLOS Computational Biology from Public Library of Science
Bibliographic data for series maintained by ploscompbiol ().

 
Page updated 2026-11-15
Handle: RePEc:plo:pcbi00:1014956