Nicolai Ree
Research Scientist, Gubra
AI/Machine Learning Worshop
Abstract
Peptides have emerged as an effective therapeutic modality for a broad spectrum of indications, including obesity. From a drug discovery perspective peptides offer attractive properties such as high receptor potency and selectivity, as well as a potential for long circulating half-life. Historically, the development of peptide-based drugs has been based on endogenous hormones; however, this approach is not applicable to targets without an identified native peptide ligand.
To address this challenge, we have implemented an AI-based computational pipeline for the de novo design of peptide binders. This pipeline allows for the specification of hotspot residues on the target protein and generates peptide sequences by maximizing hotspot proximity and employing structural metrics from AlphaFold2 [1] and proteinMPNN [2]. The predicted peptide binders show high structural diversity and virtually no sequence identity to known sequences.
We experimentally validated our AI-based strategy by synthesizing and functionally characterizing 190 de novo designed peptides targeting the active site of a G protein-coupled receptor (GPCR). Functional assays revealed receptor activity for 74% of these peptides, with the most potent hits exhibiting an IC₅₀ of 34 nM (antagonist) and an EC₅₀ of 3.2 μM (agonist). Subsequently, we optimized these de novo hits through deep mutational scans and machine learning (ML)-guided structure-activity relationship (SAR) analyses,[3] and identified several mutations that increased potency.
In summary, this study experimentally validates an AI-based pipeline for the de novo design of peptide therapeutics and demonstrates the application of ML-guided optimization to enhance drug properties. The results serve as a proof of concept for the integration of AI methodologies in peptide drug discovery, enabling the development of novel therapeutics towards previously undruggable targets in a cost-effective and time-efficient manner.
References:
[1] J. Jumper et al., Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021). DOI:10.1038/s41586-021-03819-2
[2] J. Dauparas et al., Robust deep learning–based protein sequence design using ProteinMPNN. Science 378, 49-56 (2022). DOI:10.1126/science.add2187
[3] J.C. Nielsen et al., Machine-learning-guided peptide drug discovery: Development of GLP-1 receptor agonists with improved drug properties. J. Med. Chem., 67 (14), 11814–11826 (2024). DOI:10.1021/acs.jmedchem.4c00417
Bio
Research Scientist in Computational Drug Discovery at Gubra working on peptide-based drug discovery with a focus on peptide design, screening and ML-driven analysis of large peptide libraries. Gubra is a CRO and biotech company specialized in high-end preclinical contract research and peptide-based drug discovery within metabolic and fibrotic diseases.
Prior to Gubra: PhD and later postdoctoral researcher at Bayer AG and Department of Chemistry, University of Copenhagen. This work focused on bridging quantum chemistry calculations and machine learning for molecular property predictions, in silico molecular design and computational retrosynthetic planning, especially focusing on regioselectivity and chemical stability.