Peter Diamandis Describes How Applying AI to Drug Discovery is Causing Positive Disruption to Biopharma

In a recent email blast from the desk of Peter Diamandis, he writes about developments in healthcare that could be revolutionary for the biopharmaceutical industry in creating new treatments for some of our most intractable diseases.

A small pharmaceutical firm, Insilico Medicine, has used artificial intelligence (AI) to identify a drug treatment for pulmonary fibrosis, a disease of the lungs that leads to scarring and progressively diminished breathing capacity. Known causes are adverse drug and radiation treatment reactions, autoimmune diseases, environmental triggers such as mould, and occupational exposure to particulate polluters such as asbestos, coal, and silica. You may be familiar with Mesothelioma which is asbestos-induced pulmonary fibrosis, and of course, COVID-19 which is also causing this condition in many who are infected by the virus.

What’s exciting about the work Insilico is doing is its revolutionary use of AI to identify a drug to treat pulmonary fibrosis in under 18 months, and at a cost of $2.7 million. Instead of having to do a clinical trial with 5,000, Insilico has discovered a promising treatment with 50 using its AI method called generative adversarial networks (GANS).

The content of Peter’s email blast has been edited by me (can’t help myself). But I believe I have captured its essence and hope you enjoy the retelling. As always your comments are welcomed.

What is Insilico Medicine?

Insilico Medicine is a pioneering drug company that is powered by a “drug discovery engine” that sifts through millions of data samples to determine the signature biological characteristics of specific diseases. It then identifies the most promising treatment targets and uses a new AI technique called generative adversarial networks (GANs) to create molecules perfectly suited against them.

On average, taking a drug through to FDA approval is very costly—in terms of money and time. A conservative estimate is it costs $1.8 billion and takes roughly ten years. Not only that, failure rates in preclinical target discovery are very high. And even after the targets are validated in animal models, over half of Phase 2 clinical trials fail due to the choice of target. That’s why computer scientist-turned-biophysicist Alex Zhavoronkov asked:

“What if AI could find better drug targets, faster? And what if it could create new drugs that might work on these targets for us?”

The answer became Insilico Medicine.

How Does Insilico Medicine Work?

In 2013, Zhavoronkov, passionate about longevity and anti-ageing, wrote his first book, “The Ageless Generation.” He was a researcher at Johns Hopkins University when he realized the process of drug discovery, especially in ageing-related diseases, was extremely slow and inefficient. He wondered if he could use generative adversarial networks (GANs), a subset of deep learning, to accelerate the drug discovery process.

GANs work by pitting a neural network that generates fake examples against a second neural network that tries to identify fake examples. The two programs compete until the example generator starts fooling the identifier—which means the examples become as good as “real.”

Zhavoronkov applied GANs to pharmacology to discover new pathway targets and drugs, and even predict how the latter might work before beginning costly and time-consuming clinical trials.

“Think of it as AI imagination and AI strategy,” Zhavoronkov says, first using AI to mine millions of data samples to determine the signature biological characteristics of specific diseases and signalling pathways, then instructing the AI to look at compounds that inhibit the disease with minimal side effects for humans. The AI analyzes and imagines what existing drugs on the market could be used for purposes other than their original disease targets. It then identifies the right biomarkers with minor tweaks to the molecular structure to repurpose it for another disease. If the AI can’t find a drug to repurpose, Zhavoronokov can use the same technology to imagine an entirely new rug constructing it from scratch. How precise is the technology? The Insilico drug generator can produce drug candidates that press all the right buttons in our biology. And finally, the process includes using the AI to mine massive clinical data sets and predict outcomes of clinical trials in phase 2 to phase 3 transitions, identifying the efficacy of the drug candidate.

This forms the basis of the company’s end-to-end AI platform called Pharma.AI. The components are a target discovery engine, a multiomics data analysis engine PandaOmics, a de novo molecular design engine Chemistry42, and a clinical trial outcomes prediction engine InClinico.

Drug Discovery Faster Than You Think

This year, the world has had the chance to see Insilico’s powerful, frictionless drug discovery system in action. The company has made history by identifying a preclinical drug candidate with a novel target, novel biology, for $2.7 million in 18 months. This was unprecedented, costing nearly 99% less than other pharmaceutical companies and taking a fraction of the time compared to industry norms.

The team tackled Idiopathic Pulmonary Fibrosis (IPF). Idiopathic means the disease has no known causes. Eventually, IPF progresses to potentially life-threatening pulmonary failure with currently no cure.

Insilico went to work by instructing its AI to find antifibrotic targets important in regulating fibrosis pathways and ageing. With this, the AI discovered a new target and designed ISM001-055, a small molecule inhibitor. The system predicted its promising efficacy for pulmonary fibrosis, and a good safety profile was shown. This year, the company began clinical trials in Australia, reaching a historic first in discovering a novel antifibrotic target with AI, designing it with AI, and reaching clinical trial.

Zhavoronkov envisions the development of an accelerated therapeutic pipeline to produce novel drugs in 3-5 years, depending on the disease. The company’s fast drug discovery infrastructure will be revolutionary with the technology already showing it can overcome the 99% odds of preclinical failure that besets the ponderously slow biopharma industry today.

Editor’s Note: Running counter to this narrative has been the rise of mRNA vaccine treatments for COVID-19 and the many variants that keep upping the game readiness of the pharmaceutical industry. Big pharma has shown an ability to adapt to faster timelines when faced with the current pandemic. We should be grateful for these companies that have been reinventing themselves, streamlining their discovery processes, and testing protocols to fight the virus that so far has infected close to 290 million and caused nearly 5.5 million deaths globaly.