Recruiting artificial intelligence in battle against COVID-19 and future pandemics – Guardian

Developing an effective vaccine for the current pandemic as well as treatment options for COVID-19 patients is easier said than done. The drug discovery and development processes are by no means a walk through the park, even after a potential lead is identified, countless hurdles still need to be overcome before any drug makes it to the public.

Traditional new drug development is an expensive process and is comprised of a discovery phase that includes target-based drug screening and optimization, among other processes to identify candidates to advance towards further development. Subsequent drug development involves drug combination design with these candidates, and clinical trials. Unfortunately, success rates are very low, said Dean Ho of the National University of Singapore. Ho and his collaborator, Professor Xianting Ding of Shanghai Jiao Tong University, have turned to AI to solve this problem.

For rapidly spreading pathogens with unpredictable clinical courses, [such as the current SARS-CoV-2 outbreak], this process takes too long, even with the assistance of emerging technologies, added Ho.

Even re-purposing known drugs for combination therapies can be quite challenging as choosing the right combination as well as dosage precludes the optimization of treatment outcomes, said Ho. This also limits how many drugs can be simultaneously explored, as conventional drug screening protocols cannot cope with the large data pools that get generated as a result. Given this challenge, traditional new drug development and traditional repurposing are inherently sub-optimal, he added.

In a recent paper published in Advanced Therapeutics, the team developed an AI-based platform called Project IDentif. AI that was shown to quickly screen and identify viable combination therapies for the past, present, and future infections.

Using traditional re-purposing to address SARS or MERS would be very challenging due to the aforementioned issues, said Ding. However, with a platform like IDentif.AI, combination therapy optimisation could be accomplished within days. IDentif.AI is a disease-agnostic platform. As such, it does not have to be reprogrammed, and can be immediately deployed against any novel or established pathogen.

According to the team, the core importance of IDentif.AI is that it simultaneously reconciles the optimal drugs and doses against virtually any disease model from the aforementioned extraordinarily large drug/dose parameter spaces. When good drugs are given at the wrong dose, there may be no treatment efficacy at all, said Ho. At the same time, drug dosing may also have a role in determining which drugs belong in a combination in the first place. Therefore, simultaneously pinpointing the right drugs and doses is absolutely essential.

To run a search, a small set of pre-designed combinations of drugs is given to provide a sample of the drug-dose parameter space. Imagine filling up an entire room with tiny marbles, with each marble representing a possible drug-dose combination. Our job is to find ranked list of best to worst marbles from a room filled with billions of them. This pre-designed set of combinations doesnt pinpoint every single one of them, but at least samples enough of the space to guide us to where the best one is and tells us the drugs/dosages of that optimal combination, explained Ding.

After this first set of experiments is done and the full drug-dose space is essentially mapped out for us, IDentif.AI operates off the concept that drugs and doses (inputs) are related to treatment outcomes (e.g., antiviral activity, drug toxicity) using a smooth quadratic surface (resembling a smooth mountain with one peak), added Ho. This surface is calibrated and mapped out by these set of unique initial experiments such as preventing a virus from infecting a healthy cell or shrinking a tumour (maximizing efficacy), or preventing healthy cell death (minimizing toxicity), etc.).

The map is therefore unique to every study, using different drugs and disease models, and can represent a population of cells, animals, people, or even a single patient, says Ding. To develop a population-optimized regimen, we can take biological samples pooled from a population of patients. This pooled sample can be run against a standardized cell infection model and within days, a combination will be derived. The surface map will be based on the viral/infected cell population represented by a large population of patients.

For a personalized case, if there is a patient with a high viral load, we can run the test using only their own sample, and within days, we can develop a regimen just for that patient, and the surface map will be represented by only their own sample.

And not every single drug combination needs to be screened, as once the team runs a threshold number of experiments, the map can be created and used to guide the team through the rankings of best to worst combinations based on optimal inhibition of infection and minimal toxicity.

What is really neat about IDentif.AI its ability to interrogate such as huge drug-dose space, which has already directly led to successful clinical outcomes and other indications, said Ho. As proof of concept, the team was able to identify an effective combination therapy that successfully inhibited A549 lung cell infection by the vesicular stomatitis virus (VSV) within three days of project. This compared to the months or even years that conventional drug discovery searches require, which are still only capable of exploring a tiny chemical space with poor clinical outcomes, demonstrates this technologys critical importance.

The reality is that the world will be confronted with challenges such as COVID-19 again, said Ho. We simply dont have the time or resources to wait for vaccines or antibody therapy every time. Lessons learned from COVID-19 have shown us that we cannot continue to relinquish valuable time in identifying optimal repurposed combinations. This will not solve the problem and will lead to drug shortages when some could have in fact been used correctly if a systematic optimization process was conducted.

IDentif.AI is also being adapted for additional pathogens such as Dengue fever and even the possibility of a SARS-CoV-2 mutation. If it mutates to a stage where a novel combination will be needed, IDentif.AI will be prepared to rapidly respond, said Ho.

Our aim is to give Project IDentif.AI to the world so that the next epidemic can potentially be contained or prevented using rapidly optimized drug repurposing. Implementing IDentif.AI is remarkable, rapid, and economical. As such, our work has involved healthcare economics, global health security, and surveillance experts to help us develop strategies to scale this towards widespread use on a cost-neutral basis.

Visit link:
Recruiting artificial intelligence in battle against COVID-19 and future pandemics - Guardian