The use of peptide dual agonists of the glucagon receptor (GCGR) and the glucagon-like peptide-1 receptor (GLP-1R) is one of the treatment options being researched and evaluated clinically for diabetes, obesity, and related conditions. This week, Puszkarska et al. reported machine learning (ML) models trained to predict the dual activity and potency of novel designed peptides on the two receptors.
Puszkarska and colleagues leveraged available experimental data to train ML models on the peptide sequence-function relationships to predict and identify novel multi-specific peptides. Experimental data on 125 Glucagon and GLP-1 peptide analogues were used. The EC50 values of the peptides were taken as a measure of their activation capability. An ensemble of multi-task convolutional neural networks was built to simultaneously predict peptide activity on both GCGR and GLP-1R.
Glucagon (GCG) and glucagon-like peptide-1 (GLP-1) are two helical peptides, among others, that function in the maintenance of metabolic homeostasis. They achieve this by binding to and activating the human GCGR and GLP-1R, respectively. There is a body of evidence demonstrating that GLP-1R agonists lower blood glucose and reduce obesity by inhibiting food intake. Thus, chemical analogues of GLP-1R have been approved as treatments. Also, peptide analogues with dual agonist activity on GLP-1R and GCGR are in clinical evaluation for the treatment of type 2 diabetes and obesity.
Similar dual-acting molecules like tirzepatide and maridebart cafraglutide (MariTide or AMG133) highlight the potential of such approaches. Tirzepatide is a dual agonist of GLP-1R and GIP (gastric inhibitory polypeptide) receptors, showing promising results in improving glycemic control and promoting weight loss. AMG133 is an antibody peptide conjugate with both agonist and antagonist activities, targeting GLP-1R and GIPR, aiming to achieve enhanced metabolic benefits through dual receptor modulation.
Despite this progress, there is limited understanding of sequence-function relationships of the peptides. This makes the design of new peptides with suitable selectivity profiles a slow and expensive task, relying heavily more or less on trial-and-error approaches.
Using a model-guided peptide optimization protocol, new peptides were designed with specific target activity. Three sets of five peptides were designed and tested, each set for selectivity potency on GCGR, selectivity potency on GLP-1R, or high-potency on both. Peptides P1–P3 showed potency on both GCGR and GLP-1R, and P4 and P5 have an EC50 of 68pM or better at both receptors. In addition, P1, P2, and P3 showed 7-fold higher potency than known dual agonists. The best peptide, P3, demonstrated 7.2-fold and 8.3-fold potency improvements on GCGR and GLP-1R, respectively. Several other peptides were designed with pM EC50 and selectivity for GLP-1R.
Paper: https://www.nature.com/articles/s41557-024-01532-x
Code: https://github.com/amp91/PeptideModels
