Creating Better Cyclic Peptide Inhibitors Using Machine Learning
Author Information
Author(s): Ylenia Mazzocato, Nicola Frasson, Matthew Sample, Cristian Fregonese, Angela Pavan, Alberto Caregnato, Marta Simeoni, Alessandro Scarso, Laura Cendron, Petr Šulc, Alessandro Angelini
Primary Institution: Ca’ Foscari University of Venice
Hypothesis
Can the potency of cyclic peptide inhibitors be enhanced using a combination of statistical and computational methods even with small data sets?
Conclusion
The study successfully demonstrated that combining statistical and computational approaches can rapidly generate cyclic peptide inhibitors with enhanced potency.
Supporting Evidence
- The in silico-evolved cyclic peptides showed better potency than previously developed inhibitors.
- Machine learning models were able to predict the binding affinities of new peptide sequences.
- Statistical methods combined with computational approaches led to the generation of novel peptide sequences.
Takeaway
Scientists used computers to help design better medicines called cyclic peptides, which can fight cancer more effectively, even when they didn't have a lot of data to work with.
Methodology
The study combined Random Forest Regression with Direct Coupling Analysis and Monte Carlo simulation to design cyclic peptide inhibitors.
Potential Biases
The reliance on a small data set increases the risk of selecting false positives.
Limitations
The small size of the training data set limited the predictive performance of the models.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website