Multidrug resistant (MDR) bacteria represent a global public health threat against which new drugs are urgently needed, however research towards new antibiotics is almost non-existent. Solving the antibiotics crisis is critical for society because MDR bacteria are becoming one of the main causes of death in hospitals. Antimicrobial peptides (AMPs), mostly derived from naturally occurring linear or cyclic peptides, can contribute to solving the problem, however AMPs with optimal activity and toxicity profiles are difficult to identify. The objective of SPACE4AMPS was to develop new computational tools to explore chemical space in search for novel AMPs with optimal properties, and to synthesize and test these new AMPs to identify new antibiotics for clinical development. The aim of SPACE4AMPS was not only to discover new antibiotics, but also to develop and test new methods for exploring chemical space that can make drug discovery faster and easier.
During the project, we developed computational tools to expand the diversity of AMPs. We worked on new methods to represent molecules applicable to AMPs and including a representation of chirality (left-handedness versus right-handedness). Chirality is a key property of biomolecules such as amino acids, which are the constituting building blocks of AMPs. We used these methods to obtain interactive visualizations of large datasets of AMPs and natural products. Furthermore, we developed an algorithm, called PDGA (Peptide Design Genetic Algorithm), to enable the navigation of an extremely large chemical space of 10E60 compounds.
In the laboratory, we used our computational tools to investigate structure-activity relationships in AMPs and to identify novel derivatives of the clinical antibiotic polymyxin B. These derivatives are entirely non-toxic and act by diffusing through the bacterial membrane. In a related approach, we used our computational tools to optimize the natural AMP oncocin, which kills bacteria by binding to the ribosome and inhibiting the translation of mRNA to proteins. We obtained new oncocins acting against the pathogenic bacterium Acinetobacter baumannii, against which oncocin itself is inactive. Finally, we discovered how to expand and optimize the activity and selectivity of AMPs by using machine learning to diversify the chirality of their amino acid constituents. By searching through thousands of possible chirality combinations using machine learning, we can modulate the conformation of AMPs and their bioactivity profile. A proof-of-concept was realized with a short AMP discovered in our group, as well as with the natural AMP indolicidin. This method is ground-breaking and enables optimizing therapeutic peptides for both antibiotic and general medicinal chemistry use.
Our computational work has had strong impact in the computer-aided drug design community, where our various tools have been noted and are being implemented in academic and industrial research groups. Our proof-of-concept studies with AMPs provide important experimental validations. Our work on the machine-learning guided diversification of peptide chirality is entirely innovative and opens a new and broadly applicable method to modulate therapeutic peptides. It will find its place in the rapidly growing field of machine-learning driven development of therapeutic peptides.
