rethinkPeptides Search
Menu
Study breakdown

Using Computers to Design Cyclic Peptide Drugs That Can Hit 'Undruggable' Targets

evidence
The takeaway

Computational virtual screening approaches are enabling rational design of cyclic peptides that can modulate protein-protein interactions — targets traditionally considered undruggable.

Targeting the 'undruggable'

Cyclic peptides' constrained shape enables computational screening for modulating protein-protein interactions that small molecule drugs cannot effectively target

What the researchers found

The chapter covers several computational strategies for cyclic peptide drug design:

- Virtual screening of cyclic peptide libraries can exploit their constrained shape, which is more predictable than flexible linear peptides

- Cyclic peptides, while not traditionally 'druglike' by standard rules, may achieve membrane permeability and proteolytic resistance — overcoming the two main barriers to oral peptide delivery

- Computational approaches include: diverse combinatorial virtual library generation, incorporation of various cyclization strategies and structural modifications, evolutionary algorithms for screening large libraries, machine learning approaches, and bioinformatics-guided library design

- The constrained conformation of cyclic peptides makes them more amenable to structure-based virtual screening than linear peptides

Why it matters

Protein-protein interactions drive many diseases but are considered 'undruggable' by conventional small molecules. Cyclic peptides occupy a sweet spot between small molecules and large biologics — big enough to block protein-protein surfaces but potentially small enough for oral delivery. Computer-aided design could unlock this therapeutic class at scale.

How the study worked

This is a methods chapter/review describing computational strategies for designing cyclic peptides, covering virtual screening approaches, library generation methods, cyclization strategies, and advanced techniques like evolutionary algorithms and machine learning.

What this study cannot tell us

Written in 2015, this chapter predates major advances in deep learning and AI-driven drug design that have since revolutionized the field. Many computational approaches were still in early stages with limited validated successes. The gap between virtual screening hits and actual drug candidates remains large. Oral bioavailability predictions for cyclic peptides remain challenging.

How to read the evidence

This is a methods chapter providing a review of computational approaches rather than presenting new experimental data. It describes the state of the art as of 2015 and provides practical guidance for researchers.

When this study was published

Published in 2015, this chapter captures the computational peptide design landscape before the AI/deep learning revolution. While the fundamental concepts remain valid, more powerful tools are now available.

The bigger picture

Cyclic peptides are experiencing a renaissance in drug discovery, with several FDA-approved examples (like cyclosporine) and many more in clinical development. The computational approaches described here have become increasingly powerful with advances in AI and molecular simulation, making rational cyclic peptide design more feasible than ever.

Questions still open

  • How have deep learning approaches improved cyclic peptide design since this chapter was written?
  • Can computational methods reliably predict which cyclic peptides will achieve oral bioavailability?
  • What success rate can be expected from virtual screening campaigns for cyclic peptide protein-protein interaction modulators?

Common questions

What are cyclic peptides and why are they special for drug design?
Cyclic peptides are short chains of amino acids joined into a ring shape. This circular structure makes them more stable than linear peptides (resistant to being broken down by enzymes) and can allow them to cross cell membranes. Their size — between small molecule drugs and large antibodies — lets them block protein-protein interactions that other drugs can't reach.
How do computers help design peptide drugs?
Computers can generate millions of virtual cyclic peptide structures and test how well each one fits a target protein surface — much faster and cheaper than making each one in the lab. Machine learning algorithms can learn from known bioactive peptides to predict which new designs will be most effective, dramatically speeding up the drug discovery process.

Read the original research

Computational approaches to developing short cyclic peptide modulators of protein-protein interactions.

Methods in molecular biology (Clifton, N.J.), 1268, 241-71

Citation

Duffy, Fergal J; Devocelle, Marc; Shields, Denis C. (2015). Computational approaches to developing short cyclic peptide modulators of protein-protein interactions.. Methods in molecular biology (Clifton, N.J.), 1268, 241-71. https://doi.org/10.1007/978-1-4939-2285-7_11