The new company, called Efficacy AI, uses an algorithm called MedicascyAI, or the Medicascy Artificial Intelligence Solutions Platform. The platform was developed over a period of years by computational biologist Jeffrey Skolnick, PhD, at Georgia Tech using in part NIH funding. Skolnick serves as the chairman of the scientific board and chief innovation officer for the new company.
More rapid and complete way to evaluate potential reactions
“I started out to develop a program to try to understand how proteins work to find ways to develop treatments for intractable diseases,” Skolnick told NutraIngredients-USA.
With the huge number of potential reactions that occur within the body, this could be the work of decades using traditional search methods. And that search could rely heavily on the particular gifts and insight of the searcher, which could leave a lot up to chance, Skolnick said.
The platform he developed as an alternative relies on huge data sets about protein structures and the way they interact. With that bedrock, the program can then analyze how another molecule, whether it might be a novel, protein-based drug or a chemical constituent of a natural product, could interact with the protein or proteins of interest. It’s a way to quickly unlock additional applications for these substances that might takes years to discover otherwise, if they were discovered at all, Skolnick said.
“We are working with the intersection of natural compounds with proteins which to a large extent are the workhorses of the body,” Skolnick said.
Continuous learning
There have been a number of companies in the dietary supplement industry that have tossed around the terms ‘artificial intelligence’ and ‘machine learning’ in recent years without a lot of transparency about what they really mean. Could these pronouncements in some cases just be a synonym for ‘really cool and up to date?’
For Skolnick, in order for a computer program to enter into the realm of true artificial intelligence, it has to be able to make logical connections and come to results that are outside the realm of its initial inputs. It has to be able then to execute new operations that were not within its foundational coding instructions. Skolnick said the MedicascyAI platform can do this, and can even generate what might be called hunches, based on the program’s judgment of its confidence in that particular outcome.
Skolnick said his research program has validated the predictions made by the platform. The approach is based on the generation of so-called confidence indices, which indicate a range of whether a predicted interaction is rock solid or if the prediction can be ignored.
“We’ve already analyzed 1,475 GRAS molecules and approximately 300,000 natural compounds,” he said.
He said at the moment MedicascyAI’s algorithms cover about 97% of all known proteins and make reliable predictions for about 40% of molecules found in natural products.
Partnering with natural products companies
Anthony Bellezza, CEO of EfficacyAI, said his company’s service can add confidence and speed to the development of new natural products. The program can quickly identify what a new molecule might be good for, and might unlock new applications for existing products. The program can rapidly assess the potential of new combination products and it can quantify the risk that any problems that might impede that development, such as unacceptable side effects or unwanted interactions with other natural compounds or with drugs.
“We are working with a number of well respected people in the nutraceutical and cannabis industries,” said Bellezza, who is a former senior vice president with the Rite Aid pharmacy chain.
“They are interested in how this can speed up the time to get the right product mix faster,” he said.
Bellezza said Efficacy AI, which debuted as a company in January 2020, will offer it’s program on a fee-for-service basis, but the preferred method of working with companies would be on a license or royalty fee basis figured on the amount of the products sold that were developed via the platform.