A Route to Design Novel Functional Peptides by Applying a Denoising Diffusional Model to mRNA Display Libraries.

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• In vitro directed evolution, including mRNA display, facilitates peptide ligand discovery but is limited by genetic code biases and experimental constraints, restricting exploration of sequence space.
• Targeting B-cell lymphoma extra-large (Bcl-xL), a crucial protein in cancer, is vital for therapeutic development.

Purpose of the Study:

• To apply denoising diffusion implicit models (DDIMs) for generating novel peptide ligands with high binding affinity against the cancer target Bcl-xL.
• To overcome the limitations of traditional directed evolution methods in exploring vast sequence spaces.

Main Methods:

• Trained a DDIM using high-throughput sequencing data from prior selections against Bcl-xL.
• Generated novel peptide sequences computationally using the trained DDIM.
• Experimentally validated the binding affinity, kinetics, and functionality of the generated peptide ligands.

Main Results:

• The DDIM successfully generated novel peptide sequences with high affinity for Bcl-xL.
• Most generated sequences exhibited functional equivalence to original library members, with comparable binding kinetics and affinity.
• The method identified rare sequences not readily accessible through mutation or directed evolution, efficiently exploring underrepresented sequence space.

Conclusions:

• DDIMs serve as a powerful complement to directed evolution, enhancing the exploration of sequence space for ligand discovery.
• This computational approach accelerates the identification and optimization of molecular properties for diverse therapeutic targets.
• The framework offers a broadly applicable strategy for advancing peptide ligand discovery and development in oncology and beyond.