Researchers from the Laboratory of Computational Systems Biotechnology at EPFL, Switzerland, have developed a database – NICEdrug.ch – that could accelerate the process of drug development. The new workflow for drug discovery will help to identify suitable repurposing candidates for multiple diseases.
The need for new drugs
There is an urgent need for new and effective drugs not only to treat emerging diseases but also to help tackle the rising numbers of resistant bacteria. However, drug development is far from straightforward. Approximately 90% of drug candidates in clinical trials are rejected due to unexpected toxic side effects. This makes the current development pipeline inefficient and costly. Improving this system is therefore extremely important to fulfil the need for new small molecules.
Currently, high-throughput screens of drug candidates against a set of target enzymes are used to identify potential compounds. However, there are limitations with the biochemical space, which affect the utility of this method. Therefore, researchers are turning to in silico strategies to identify target compounds. Mechanistic-based approaches and machine learning (ML) are the two main in silico methods. The former does not take reactive sites or neighbouring atoms into consideration, whilst ML approaches require large and high-quality data sets that can be difficult to obtain. In addition, neither consider the effect of human biochemistry on potential drugs.
A more detailed workflow for drug discovery
To overcome the current pipeline inefficiencies, the researchers developed a database named NICEdrug.ch. It accounts for the effects of patient metabolism as well as including a more detailed analysis of drugs’ molecular structures. With this tool, the scientists were able to automatically analyse the functional, reactive, and physicochemical properties of 250,000 small molecules. The action mechanism, metabolic fate, toxicity and drug repurposing possibility could then be suggested for each compound.
The NICEdrug.ch database
Over 70,000 existing small molecules were used to build the initial NICEdrug.ch database. Those that were not found to have drug-like properties were cut, resulting in a final database of 48,544 unique small molecules.
In addition, NICEdrug.ch used information about the structure of reactive sites and their surroundings to create ‘fingerprints’. These fingerprints were shown to better predict similar reactivity between different molecules than overall molecular structure. Overall, 20 million reactive site-centric fingerprints were generated, which were able to identify reactive similarities between drug-drug and drug-metabolite pairs.
Identifying drugs for repurposing
The extensive information stored within the NICEdrug.ch database, especially the inclusion of human metabolism, allowed researchers to identify drugs that could be repurposed for multiple diseases.
One of the most commonly used anti-cancer drugs is 5-fluorouracil (5-FU). However, many patients treated with 5-FU experience toxic side effects. Using NICEdrug.ch, the researchers were able to identify the toxic processes within the drug’s metabolism. It is hoped that these findings will guide future attempts to develop treatments that alleviate 5-FU’s side effects.
NICEdrug.ch suggested over 5,000 drugs and drug candidates to target liver-stage malaria. The top candidate was identified as shikimate 3-phosphate, which is predicted to have no side effects in the human host cell. Previously, shikimate 3-phosphate has been used to treat E. coli and Streptococcus infections with minimal toxicity, adding promise to this exciting finding.
Furthermore, the study attempted to identify potential drugs against COVID-19. Overall, 1,300 molecules were found, including 465 molecules that are already used against other diseases. However, further studies will be needed to confirm that the compounds discovered are safe and effective for the treatment of COVID-19.
Future implications of the new workflow for drug discovery
The team showed that their database could accurately predict the behaviour of drug-enzyme pairs over 70% of the time. Amazingly, half of the pairs tested were 100% accurate. This performance identifies NICEdrug.ch as a potentially key tool in the future of drug identification.
The database is currently available as an open-access resource. Although future studies will be required to verify the behaviours of identified molecules, the team hopes that other scientists will make use of NICEdrug.ch to better inform and streamline their research.
Image credit: Background photo created by freepik – www.freepik.com