Researchers have developed a machine learning algorithm, called NRPminer, that makes it easier for scientists to develop drugs from natural products.
Historically, natural products have made a major contribution to pharmacotherapy. Many of the antibiotic, antifungal and antitumour medications available today were, in fact, discovered from natural products. Perhaps the most well-known drug derived from a natural product is penicillin – the first naturally occurring antibiotic that was accidentally revealed by Sir Alexander Fleming in 1928.
Nevertheless, natural products still present several challenges for drug discovery, such as technical barriers to screening, isolation, characterisation and optimisation. Therefore, the process of uncovering natural products is time consuming, labour intensive and ultimately incurs huge costs. Subsequently, there has been a decline in their pursuit by the pharmaceutical industry since the 1990’s. However, in recent years, several technological advancements have begun to address these challenges.
As a result of these scientific developments, could we be witnessing the revitalisation of using natural products as drug leads?
Non-Ribosomal Peptides
Non-Ribosomal Peptides (NRPs) are an important class of microbial natural products. They are used to make antibiotics, immunosuppressants, anticancer agents, toxins, siderophores, pigments and cytostatics. However, because NRPs are not directly encoded in the genome, their discovery remains a slow process. Instead, they are produced by metabolic pathways that are encoded by biosynthetic gene clusters (BGCs). This makes NRPs difficult to detect and even more problematic to identify as potentially useful.
Throughout the past decade, genome mining methods have become increasingly popular for predicting NRP sequences from their BGC sequences, as these methods also reveal the final molecular products that they encode.
However, existing genome mining strategies have struggled to reveal the chemical diversity of NRPs. For example, these techniques fail to correctly identify post-assembly modifications (PAMs). PAMs transform core NRPs into mature NRPs. They are responsible for making NRPs the most diverse group of natural products and play a critical role in NRPs mode to action. Therefore, until now, the potential of large-scale NRP discovery had not been realised.
NRPminer tool: A machine learning algorithm
Recently, researchers from Carnegie Mellon University’s Computational Biology Department in the School of Computer Science developed a novel process using a machine learning algorithm to identify NRPs from different environments.
The team was able to scan the metabolomics and genomic data for about 200 strains of microbes. The artificial intelligence tool, called NRPminer, was able to recognise which microbial metabolites were likely to respond to which natural products. Then, armed with this information, scientists could isolate the NRP and begin developing it for a possible drug.
Overall, the machine learning algorithm identified 180 unique NRPs, including four previously unreported ones. The products are currently being analysed by a team at the Institute for Molecular Bioscience at Goethe University in Germany, where two have already been found to have potential anti-malarial properties. This highlights the fact that applying this novel machine learning approach to large-scale metagenomic data could reinvigorate the search for drugs from natural products.
Hosein Mohimani, a lead researcher in the project, explained: “What is unique about our approach is that our technology is very sensitive. It can detect molecules with nanograms of abundance. Our hope is that we can push this forward and discover other natural drug candidates and then develop those into a phase that would be attractive to pharmaceutical companies.”
Image credit: FreePik rawpixel.com
