EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A distribution-aware and functionally relevant novel framework for generation and discovery of bioactive peptides

Rachit Abhigyan, Vikas Sood, Pooja Arora, Baljeet Kaur

Artificial Intelligence in the Life Sciences · 2026-12 · 10 · 100190

DOI: 10.1016/j.ailsci.2026.100190 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Artificial Intelligence in the Life Sciences, 2026-12. Verfasst von Rachit Abhigyan und 3 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Recent advances in artificial intelligence have unfurled new possibilities for discovering bioactive peptides by allowing exploration of large sequence spaces computationally. However, existing approaches have their limitations. Distribution-learning models may produce peptides similar to experimentally validated sequences, but these peptides may not exhibit high predicted activity. In contrast, optimization-based methods may identify sequences with high predicted activity, although these may differ from the distribution of experimentally validated peptides. To address these limitations, a two-phase framework is proposed that combines distribution learning with evolutionary optimization. In the first phase, Variational Autoencoders (VAE), Autoregressive Transformers (ART), and Token Diffusion Transformers (TDT) are used to generate distribution-informed seed peptides. Further, these peptides are used with Hill Climbing (HC) and Genetic Algorithm (GA) optimization procedures that iteratively improve performance. The proposed framework was evaluated using a dataset of experimentally validated IL-2-inducing peptides based on two separate prediction models. Autoregressive Transformer combined with Hill Climbing achieved the best overall performance, with a top-200 mean IL-2 induction confidence score of 0.8803±0.0117 while reducing KL divergence from 2.2639±0.0471 for standalone Hill Climbing to 0.8080±0.0774. For IL-13-inducing peptide dataset, ART-initialized Hill Climbing achieving the top-200 mean IL-13 induction score of 0.8927±0.0247 while reducing KL divergence from 1.7119±0.0733 to 0.5758±0.0613. In the anti-cancer peptide dataset, ART-initialized Hill Climbing achieved top-200 mean prediction score of 0.9545±0.0123 with a KL divergence of 1.3413±0.1708. Overall, the framework provides a potential approach for balancing functional optimization and distributional similarity thereby prioritizing candidate peptides for experimental validation.

Zitieren

APA

Abhigyan, R., Sood, V., Arora, P., & Kaur, B. (2026). A distribution-aware and functionally relevant novel framework for generation and discovery of bioactive peptides. Artificial Intelligence in the Life Sciences, 10, 100190. https://doi.org/10.1016/j.ailsci.2026.100190

Vancouver

Abhigyan R, Sood V, Arora P, Kaur B. A distribution-aware and functionally relevant novel framework for generation and discovery of bioactive peptides. Artificial Intelligence in the Life Sciences. 2026;10:100190. doi:10.1016/j.ailsci.2026.100190

BibTeX
@article{Abhigyan2026,
  author = {Abhigyan, Rachit and Sood, Vikas and Arora, Pooja and Kaur, Baljeet},
  title = {A distribution-aware and functionally relevant novel framework for generation and discovery of bioactive peptides},
  journal = {Artificial Intelligence in the Life Sciences},
  year = {2026},
  volume = {10},
  pages = {100190},
  doi = {10.1016/j.ailsci.2026.100190}
}
Kostenlos starten →