Insilico Medicine and the AI Revolution in Drug Discovery

Recently, the Atlantic ran a feature story on the many ways artificial intelligence (AI) is shifting the realm of possibilities for discoveries in medicine – allowing scientists to generate new hypotheses and new molecules to treat diseases and ultimately helping us crack the code on human biology. The story highlighted one drug in particular that has come to symbolize this new revolution in medicine. It’s a drug for the chronic and often fatal lung disease idiopathic pulmonary fibrosis (IPF) from global AI-driven biotech company Insilico Medicine, the first drug with an AI-discovered target that was also designed by generative AI to reach Phase II trials with patients.?

INS018_055, as the drug is officially known, is one of many AI-designed drugs in Insilico’s pipeline, but it holds a special place. It is the lead program, and it is the Company’s “moonshot” – the first validation of a rapidly evolving end-to-end software platform called Pharma.AI, built on a backbone of NVIDIA technology, that has proven that with a combination of sophisticated algorithms, reinforcement learning, vast quantities of diverse data, robotics, and human expertise, it can vastly accelerate early drug discovery and advance new therapeutics for a range of hard-to-treat diseases.?

The Evolution of Insilico’s Generative AI Platform

Insilico has made substantial leaps in technological capabilities since the Company first announced the preclinical candidate for its lead drug in Feb. 2021.?

What began in 2016 as a new approach to use machine learning to make sense of how biological pathways are activated in breast cancer in order to identify new genes that can be targeted for treatments (published in Nature Communications) has evolved into a sophisticated, interconnected, end-to-end generative AI platform for accelerating and improving early drug discovery that has been validated through many therapeutic programs, both partnered and internal, with five in clinical stages.?

Insilico’s Pharma.AI platform was developed by a team of biologists, bioinformaticians, computer scientists, and software engineers who turn reams of available scientific data into curated information that allows scientists to immediately jump in, perform analysis, generate new hypotheses, and design new, high-quality molecules from scratch using generative AI.

The platform benefits from the NVIDIA BioNeMo cloud service, which includes pre-trained AI models for drug development and supports data-driven protein design.?

“We made a bet on generative AI early, tapped NVIDIA’s best-in-class technology, worked hard, and now we can tangibly demonstrate substantial R&D performance boosts and real clinical-stage drugs imagined by AI,” says Alex Zhavoronkov, PhD, founder and co-CEO of Insilico Medicine.?

“Insilico Medicine saw starting a decade ago the great potential of AI to accelerate the entire drug discovery process,” said Kimberly Powell , vice president of healthcare at NVIDIA. “Today, we continue to partner with Insilico to incorporate NVIDIA BioNeMo and turbocharge the Pharma.AI drug discovery and development platform with the latest advances in accelerated computing and generative AI to enable Insilico and their partners’ drug discovery pipelines.”

AI for Targets, Drugs, and Clinical Trials

The platform has three main components, each with its own set of specialized tools. The Biology42 component contains PandaOmics , a tool with over 20 AI and statistical models that analyzes datasets from diseases and biological processes in the context of scientific knowledge and identifies potential targets for new drugs in minutes.?

“One? of? the? key? differences? between? PandaOmics? and? its? competitors? is? the? enormous? database? of? pre-created? datasets? and? meta analysis users? get? access to with their? subscription,” said Kyle Tretina, PhD , Alliance Manager for AI Platforms at Insilico Medicine, in a recent webinar. “Our? team? has? been? working? all? year? to? scan? the? exponentially? growing? data? that? is? publicly? available? and? then? curate? the? most? relevant? data? for? the? indications? that you’re? most? interested? in. Overall,? this? improvement? has? led? to? a? 10-fold? increase? in? the number of? pre-calculated? multi-OMICs? disease? projects? that? are? available? to? the? user? compared? to? our? last? version.”?

The latest updates to the Biology42 platform include a generative? biologics? app? that? helps? users generate? novel? peptides,? as well as the AI-powered robotics lab, Life Star, which performs experiments that allow scientists to test their hypotheses.?

Once a promising target is identified, this same platform can design a 3D molecule from scratch that has all the characteristics needed to act on the target to inhibit the disease and be safe, potent, and metabolically stable. Chemistry42 includes the Generative Chemistry tool, which works like ChatGPT in the chemical space. Instead of dreaming up new poems from existing text, or new faces from existing images, it utilizes known chemical structures to design brand-new therapeutics. These high-quality molecules, ranked by the AI platform, can be produced and tested in order to identify the most promising drug candidates. The latest version of Chemistry42 includes a Golden Cubes module for ligand-based scoring of the kinase selectivity of small molecules to minimize unintended, off-target effects that can lead to severe side effects, as well as a refinement tool that allows users to add structures to a finished experiment in order to test specific ideas aiming to improve the ligand.

Most drugs ultimately fail in later-stage clinical trials. A report from the organization BIO found that between 2006-2015, just 9.6% of drugs from Phase I clinical trials were ultimately approved. Phase II clinical trials had the highest failure rate among the four development phases – with only 30.7% of drug candidates advancing to Phase III.?

The final component of Pharma.AI – Medicine42 – addresses this roadblock directly by doing what AI does best – predicting outcomes. A tool called inClinico uses AI to find patterns within drug toxicology reports, clinical trial data, and various disease pathways in order to estimate the likelihood of success in a drug candidate transitioning from Phase II to Phase III. This tool can also model the critical parts of patients’ eligibility criteria relating to a trial’s success so users can see the impact of adding or removing different criteria on clinical trial outcome scores.

Working with pharma partners, Insilico’s Pharma.AI platform is on the frontlines of the AI-driven drug discovery revolution, which relies on the latest technology and the brightest scientific minds to solve the mysteries of diseases, aging, and biological processes. With the added benefit of robotics capable of running parallel experiments, these tools are bringing real value to an industry that has struggled with high costs and high failure rates. There are tens of thousands of diseases without cures and many more patients awaiting new cures.?