Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More
We've already seen how AI can be used to help uncover the structure of proteins and discover new materials, so it would follow naturally that a specially trained AI model could also use data of previous drug molecular structures to find new drugs and accelerate treatments for some of the toughest diseases and medical conditions out there.
That's exactly what one company, VeriSIM Life, based in San Francisco, is hoping to do, through its platform, BIOiSIM, which contains a massive data lake of more than 3 million compounds and 5,000 human and animal datasets with AI models trained atop it, allowing pharmaceutical researchers to find, develop, and test new compounds virtually before spending the money to test them in real clinical trials.
Earlier this year, VeriSIM introduced a new feature to the platform, AtlasGEN, focused on providing biological translation simulations -- estimates of how well a drug actually works on the human (or animal) body. With this information, pharma researchers can decide not only which new drugs are worth developing for real, but which kinds of animals to test them on.
"We can reduce animal testing," said VeriSIM Life CEO and founder Dr. Jo Varshney, in a video call interview with VentureBeat several months ago. "We can refine those animal experiments so you don't have to do it in a cohort of 50 animals, for example. You can only test your drug in a few and validate it more computationally."
Not only is this more humane for animals, but it also saves researchers time and money by preventing them from testing drugs in animal subjects whose biology, physiology and genetics mean that the drug being researched won't actually effect them in the way researchers hope.
A personal quest for better drug research tools and technology
For Varshney, the quest is a personal one: not only was her father in the pharmaceutical business, but she has been interested in the field since she was two years old, she told VentureBeat, and originally began her career as a veterinarian before obtaining a PhD in genomics and computational sciences at University of California San Francisco.
"I spent a lot of time before starting the company, back in the day, on machine learning -- supervised, unsupervised -- and asking, 'if we have enough knowledge and understanding of medicine, biology, chemistry, can we use that knowledge and make predictions for novel knowledge, novel chemistry, novel molecules?'" Varshney said.
Already, VeriSIM Life has helped four clients bring their drugs to clinical trials, according to Varshney.
The company has raised millions from the likes of Intel Capital, Village Global, Susa Ventures, Stage Venture Partners, Loup Ventures, and Twin.
The problem with current drug research and discovery: it's expensive and has a high failure rate
The global pharmaceutical industry is worth around $1.6 trillion as of the latest figures (according to market research firm Statista), yet the amount spent on research and development of new drugs has ballooned 10X in the U.S. alone (adjusted for inflation) since the 1980s, according to the Congressional Budget Office.
Drug research also has high failure rate: an estimated 90% of all drugs fail clinical trials, according to one study from 2016, even though the average drug requires more than $870 million to develop and takes around 10-12 years per drug!
VeriSIM estimates its BIOiSIM platform can cut the timeframe to go from research and development to requesting an authorization from the FDA to run clinical trails (Investigational New Drug or IND Application) by 2.5 years.
It also claims to offer 82% greater accuracy at modeling drug effects than other "non-AI" methods.
What's under the hood of VeriSIM Life's BIOiSIM and AtlasGEN?
VeriSIM Life's BIOiSIM platform is a sophisticated computer program consisting of multiple AI models and datasets, according to Varshney.
"We use AI from machine learning methods, generative adversarial networks (GANs), generative AI, to actually identify new molecules in a massive space of 10 to the power of 63 and then distill it down to the best molecular structure," the CEO and founder told VentureBeat.
But equally important to the platform is that VeriSIM Life has also created virtual analogs of real world species, including humans, dogs, rats, pigs, and numerous other animals regularly used in clinical trials to test drug compounds.
"We combined the knowledge from chemistry, physiology, the different types of animals that are being used in testing, and codified all that, and then the different types of patients-based 'omics' data -- so think genomics, proteomics and such -- and distill it all down into a score, which is called the Translational Index. It's inspired by a credit score."
That score, delivered as a rating between 1 and 10, with 10 being the most efficacious and 1 being the least, allows pharmaceutical researchers to judge in advance whether or not a drug is worth pursuing into clinical trials, and just as important -- which animal models to test it on to obtain the desired results.
So, for example, if researchers wanted to test a new cholesterol reduction drug -- they could use VeriSIM LIfe's BiOSIM AtlasGEN feature to research the best compounds to use, and then the Translational Index to get scores for which animals would be best to test it on as well as whether or not the drug would perform well in humans, providing them with the best approach to focus their efforts for success.
"Even if it's efficacious in animals, but if it's not efficacious in human, the score goes down," Varshney told VentureBeat.
Altogether, BIOiSIM and AtlasGEN can run more than 800 billion different scenarios, according to VeriSIM's website.
Furthermore, Varshney told VentureBeat that its team of computer engineers and accredited researchers frequently go in and customize its platform and animal models for different clients depending on their specific needs.
"For example, if we know that a drug will be toxic in the liver, our experts go and build in more specific details within the liver organ model to show how that toxicity would differ between a rat and a dog and a human -- that kind of choice is where we spend a lot of time working," the CEO said.
As for monetization, Varshney said that VeriSIM life takes a percentage of drug revenue developed on the platform, but also offers subscription software-as-service based pricing annually or on a project-by-project basis.
Part of a wave of AI-driven healthcare apps and platforms
AI's infiltration in the healthcare sector isn't limited to drug discovery, of course.
We've also covered companies using gen AI models(including OpenAI's underlying GPT) to offer doctors recommendations for cancer screenings based on patient profiles; propose diagnosis; provide alerts and predictions of hospital patient health; create whole new generative apps for doctors and patients; and much more.
Clearly, VeriSIM hopes to make a dent with its BIOiSIM platform and Translational Index scores, and help drive down the costs, increase the success of drug trials, and ultimately, improve and extend the lives of people around the world.
