As the novel coronavirus began to spread worldwide, the UK government promised to "throw everything" at finding a vaccine. Imperial College London and the University of Oxford received £42.5 million ($55.75m) in funding and thousands of suppliers were primed to mass-produce a viable vaccine -- should one be proven safe and effective.
Now, clinical trials are underway in countries including the UK, US, South Korea, and Russia. In the latter case, however, the breakneck speed of tests and approval of mass-vaccination efforts has drawn criticism.
While hopes for a vaccine remain, it is crucial to note the ongoing efforts of researchers to develop therapeutics for lessening the severity of COVID-19, especially when contracted by individuals in high-risk groups or those suffering from pre-existing medical conditions.
If we can't cure COVID-19, we can, at least, try to discover or develop drugs and therapies that reduce the respiratory illnesses' impact on the most vulnerable.
Projects focused on COVID-19 may have the potential to reduce the human cost of the pandemic and provide potentially lucrative licensing revenue streams for countries where researchers find viable therapeutics for fighting the disease.
With an urgent need to find or create effective therapies -- combined with new cash injections provided by private companies or government organizations -- research teams are examining how computing power and new technologies can speed up the drug discovery process.
Computational modeling can give scientists insight into the likely success of a drug in fighting a medical condition. Therefore, it can be a crucial component in quickly bringing experimental drugs from trial stages to regulatory acceptance. However, you need to start from the beginning, with chemical composition.
Enter IBM, with the debut of IBM RoboRXN for Chemistry, a free AI service for predicting chemical reactions and the development of molecules -- and a system being used by the tech giant to discover ways to inhibit proteins associated with the novel coronavirus.
Big Blue debuted the AI and cloud-driven platform at a virtual event in Zurich, Switzerland on Wednesday, together with a demonstration of how the new technology could be used to predict and model the outcomes of molecule reactions during drug development.
When it comes to drug production, on average, IBM says it takes 10 years for a new material or drug to be discovered and reach the market, as well as at least $10 million in funding.
The aim is to cut this down to one year and only $1 million.
IBM RoboRXN for Chemistry brings together cloud, AI, and automation to tackle complex organic chemistry reactions; in particular, unknown organic chemistry reactions and synthesis that could pave the way for new drug discoveries.
Teodoro Laino, manager of IBM's Future of Computing for Accelerated Discovery, explained how it works. A chemist could be sitting at home who was willing to make a molecule, and after connecting to RoboRXN for Chemistry via a web browser, they draw the molecule.
RoboRXN would then recommend optimal scientific routes and the best starting material available commercially.
Once submitted, RoboRXN would self-program itself to "execute the process in an autonomous laboratory." In other words, experiments could be conducted remotely with the right integration and hardware setup.
In a demo, IBM researchers compared the process to cooking an apple pie. Each component -- such as the pastry -- requires a specific set of instructions.
RoboRXN can accept instructions from published literature on molecule types and reactions, simply cut-and-paste by chemists, or the system can recommend how an experiment should be performed.
Three AI models have been trained for this task: the first is focused on retrosynthetic analysis and determining the 'ingredients' -- including precursors that are commercially available -- and translating text-based descriptions into what Laino describes as a "sentence of atoms."
The second and third models focus on synthesis actions, harnessing a dataset made up of millions of chemical reactions already published in literature and patents. Processes can include concentration, stirring, purging lines via brine, temperature changes, and adding different compounds.
Once the chemical synthesis process is complete, an analytics report is automatically generated to help chemists further their research.
In tests, IBM has benchmarked a 90% accuracy rate, and while the black box conundrum -- trying to understand how an AI algorithm makes decisions -- is still a problem, the team says there is an "ongoing effort" to improve the transparency of its models' decision-making.
Laino says that the platform can accelerate material discovery and could also pivot the traditional chemistry field into a high-tech business.
As a hardware-agnostic and scalable solution, the team says that RoboRXN can be a valuable tool not just for a chemist forced to stay at home due to the pandemic, but also for large organizations.
In the future, RoboRXN may become an on-premise and private rather than public cloud solution, and potentially, the system may also be established as a 'chemistry as a service' offering to the enterprise.
COVID-19 has cast a spotlight on our existing medical research facilities, healthcare providers, and funding. The disease has also become a catalyst for additional support being offered to scientists involved in vaccines and drug discovery.
Supercomputers are being employed in the US to model how existing drugs attach to viruses, reducing research timetables from years to months. AWS, Microsoft, and Google are also working with pharmaceutical giants to create hyperscale cloud and artificial intelligence (AI) systems to make drug discovery quicker and cheaper. Modeling and predicting outcomes for new drugs through computers that can process complex data at rapid speed can reduce the labor required of research teams.
As previously reported by ZDNet, there are generally four stages of drug development in the US to secure Food and Drug Administration (FDA) approval. After passing preliminary checks, stage 2 tests can cost up to $19 million on average, rising to roughly $53 million in stage 3, whereas only 13% of drugs end up available commercially.
When trials for a single drug can prove this costly, any means to reduce expenses and the time required to develop and test new material is a boon to the medical community and pharmaceutical companies.
A live stream of the Zurich laboratory is available here. Academic papers on the AI models are also available.