Exploring and expanding the chemical multiverse of peptides

Edgar López‐López, Javiera V. Castañeda, Massyel S. Martínez-Cortés, César de la Fuente‐Núñez, José L. Medina‐FrancoChemical 2025 · DOI

Abstract

Peptides occupy a unique and rapidly expanding domain within the broader chemical space, offering exciting opportunities for therapeutic, nutritional, cosmetic, and materials applications. While efforts to characterize chemical space have traditionally focused on small molecules, growing evidence underscores the value of extending this concept to include peptide-based compounds. In this review, we survey recent advances in the exploration and expansion of the peptide chemical space, focusing on short peptides that bridge chemoinformatics and bioinformatics perspectives. We begin by briefly discussing the impact and applications of peptides in various research and industry areas and then examining the theoretical and practical size of the peptide chemical space, emphasizing how naturally occurring and synthetic peptides vastly increase its diversity. We then discuss molecular representations—from conventional notations to specialized peptide descriptors—highlighting their impact on library design, structural analyses, and activity predictions. Visualization methods and machine learning models are presented as tools for mapping structure-property and structure-function relationships. Next, we explore computational strategies for de novo peptide generation, driven by advances in generative modeling and high-throughput screening. Throughout, we emphasize the role of open-source resources and integrated computational pipelines that combine chemo- and bioinformatics approaches to enhance data quality and predictive performance. We conclude by identifying major challenges—such as the complex structural landscape of peptides, data curation, and the need for consensus screening methods—and outline emerging opportunities for further expanding and refining the peptide chemical space.

Introduction: The Combinatorial Challenge

Peptides represent a unique "multiverse" in chemical space, bridging the gap between small molecules and large proteins. While small-molecule chemical space is well-mapped, peptides present a daunting combinatorial challenge. A simple 10-residue peptide composed of the 20 standard amino acids yields $20^{10}$ (approximately 10 trillion) possible sequences. When incorporating non-canonical amino acids (ncAAs) and post-translational modifications (PTMs), the number of potential molecules becomes astronomically larger than the number of atoms in the universe. This review explores how researchers navigate and expand this vast domain to discover novel therapeutics, functional foods, and advanced materials.

Methodological Convergence: Bioinformatics meets Chemoinformatics

Traditionally, peptides were treated either as sequences (bioinformatics) or as 3D atomic arrangements (chemoinformatics). Modern research requires a hybrid approach to effectively map the "privileged" molecules that function as drugs or materials.

  • Bioinformatics (Sequence-centric): Utilizes specialized notations like HELM (Hierarchical Editing Language for Macromolecules) and PLN (Protein Line Notation) to manage complex architectures, such as cyclic peptides and branched chains.
  • Chemoinformatics (Atom-centric): Employs molecular fingerprints like MAP4 (MinHashed atom-pair) and MXFP (Macromolecule extended atom-pair) to capture fine-grained chemical environments. This is essential for predicting how a peptide interacts with a biological target.

The integration of these fields allows for advanced generative modeling (e.g., PepINVENT) and genetic algorithms to design de novo peptides with specific properties, such as high stability or membrane permeability, moving beyond what is found in nature.

Mapping the Known and Unknown: Data Resources

The review provides a comprehensive directory of over 20 peptide-related databases, which serve as the foundation for computational exploration. Key resources include:

  • General Repositories: PDB, UniProt, and PeptideAtlas.
  • Specialized Databases: DRAMP and APD6 (antimicrobial peptides), Norine (non-ribosomal peptides), and SATPdb (structurally annotated therapeutics).
  • The "Extinctome": A frontier involving peptides derived from the proteomes of extinct organisms, used to identify "ancient" antimicrobial leads. This represents a fascinating intersection of paleontology and drug discovery, where researchers mine sequences that may have been lost to evolutionary time.
  • Virtual Libraries: Massive virtual spaces, including libraries of up to $10^{60}$ peptide-like molecules generated from diverse building blocks.

Strategic Insights: Subspaces and Stereochemistry

Chemical space is not a single entity but a collection of "subspaces" defined by the specific descriptors used to view them. A peptide might appear similar to another in sequence space but look completely different in 3D electrostatic space.

Furthermore, stereochemistry (chirality) is a critical but often overlooked dimension. Moving beyond standard L-amino acids to include D-enantiomers and other chiral modifications is essential for creating peptides that resist enzymatic degradation, a major hurdle in peptide drug development. The "thin line" between bioinformatics and cheminformatics is disappearing as modern research treats peptides as both sequences of letters and 3D arrangements of atoms simultaneously.

Case Studies and Practical Applications

The synthesis of computational design and experimental validation has led to several high-impact successes:

  • Metabolic Success: The evolution of GLP-1 analogs (such as semaglutide) demonstrates how specific structural modifications can transform a peptide with a half-life of minutes into a blockbuster drug with weekly dosing.
  • Mapping the Extinctome: Using network-based visualization (t-SNE and scaffold similarity maps), researchers identified ancient peptides with potent anti-MRSA activity that lacked the hemolytic side effects common in modern analogs.
  • Focused AI Libraries: AI-driven "focused libraries" are now outperforming random high-throughput screening by pre-filtering for drug-likeness (QED) and low toxicity before synthesis.

Future Directions and Grand Challenges

Despite rapid progress, several "grand challenges" remain for the next decade:

  1. Data Curation: There is a severe lack of standardized, high-quality datasets that include non-canonical amino acids and detailed chirality information.
  2. Explainable AI: Many deep learning models for peptide activity lack interpretability. Researchers need to understand why a specific sequence is bioactive to refine designs.
  3. Synthesis Constraints: Computational design still outpaces chemical synthesis. Automated platforms must improve to incorporate complex building blocks like chiral sulfur atoms or modified backbones.
  4. Consensus Screening: There is a need for unified computational pipelines that integrate structural, physical, and biological data into a single, holistic prediction of peptide function.

Conclusion

The exploration of the peptide chemical multiverse is transitioning from random discovery to intentional design. By bridging the gap between sequence and structure, and leveraging the power of generative AI and curated databases, the next generation of peptide-based solutions is within reach.

Summary updated

Abstract

Peptides occupy a unique and rapidly expanding domain within the broader chemical space, offering exciting opportunities for therapeutic, nutritional, cosmetic, and materials applications. While efforts to characterize chemical space have traditionally focused on small molecules, growing evidence underscores the value of extending this concept to include peptide-based compounds. In this review, we survey recent advances in the exploration and expansion of the peptide chemical space, focusing on short peptides that bridge chemoinformatics and bioinformatics perspectives. We begin by briefly discussing the impact and applications of peptides in various research and industry areas and then examining the theoretical and practical size of the peptide chemical space, emphasizing how naturally occurring and synthetic peptides vastly increase its diversity. We then discuss molecular representations-from conventional notations to specialized peptide descriptors-highlighting their impact on library design, structural analyses, and activity predictions. Visualization methods and machine learning models are presented as tools for mapping structure-property and structure-function relationships. Next, we explore computational strategies for <i>de novo</i> peptide generation, driven by advances in generative modeling and high-throughput screening. Throughout, we emphasize the role of open-source resources and integrated computational pipelines that combine chemo- and bioinformatics approaches to enhance data quality and predictive performance. We conclude by identifying major challenges-such as the complex structural landscape of peptides, data curation, and the need for consensus screening methods-and outline emerging opportunities for further expanding and refining the peptide chemical space.
BibTeX
@inproceedings{lopezlopez2025exploring,
  author = {Edgar López‐López and Javiera V. Castañeda and Massyel S. Martínez-Cortés and César de la Fuente‐Núñez and José L. Medina‐Franco},
  title = {{Exploring and expanding the chemical multiverse of peptides}},
  booktitle = {Chemical},
  year = {2025},
  doi = {10.1039/d5sc04465k},
}
Figure