Meet the scientists using generative AI to cure cancer

The company is now on the cusp of a revolution in drug discovery with the use of generative AI which can not only cut down the time it takes to process cancer medicines, but save a lot of money while doing it.

“The whole [drug discovery] process is extremely inefficient, slow and expensive so it takes over 10 years to develop a drug,” said Alex Aliper, cofounder and president of Insilico Medicine, in an exclusive interview with Arabian Business.

Currently, it takes between 10 to 12 years and costs almost $2 billion to put a drug on the market, a major issue Big Pharma has been grappling with for many years. However, Insilico’s Pharma.AI, an end-to-end drug discovery platform, accelerates the process through disease target identification, the generation of novel molecules data, and predicting clinical trial outcomes.

The AI-powered platform comprises PandaOmics, Chemistry42, and inClinico – the three components that carry out the drug discovery process through the use of millions of data samples and multiple data types.

Aliper believes that the drug discovery engine holds the promise of discovering drugs “as efficient and cheap as possible.”

The development of the Pharma.AI platform is now being supervised from Abu Dhabi, making it the first project of its kind in the UAE. This is a seemingly incredible feat as no drug has ever been discovered in the Middle East region.

Dialling in with Aliper from Chengdu, China was the company’s founder and CEO, Dr. Alex Zhavoronkov, who said he “hopes to change that.”

There are currently over 10,000 unmet medical needs in the world without a treatment option. This is driving the rush for more advanced drug discovery – and what better method to do so than generative AI?

“Only with the power of AI are we able to kind of crack this problem as we see that conventional methods are failing in Pharma,” Zhavoronkov said.

Insilico’s AI platform identifies the right target – a protein or an entity within an organism – for the disease of interest. The target is then duplicated in order to modulate the disease and either “reverse it or ameliorate it,” he said. Once identified, it designs the small molecule or a molecule to modulate a target to cure the disease, making way for lead optimisation to come into play. Following this, the molecule with the highest potential for success is selected and then developed for clinical trial.

“We have many areas where AI provides significant speed-up in efficiency to the process from disease hypothesis to target identification to lead optimisation and selection of a candidate. To do that, we have built the Pharma.AI platform,” the CEO added.

The platform integrates three products: PandaOmics, which covers target identification – the first part of the discovery process; Chemistry42, which discovers novel molecules with modulator target identified in PandaOmics; and inClinico, a tool specifically designed to predict the likelihood of success in a phase 2 clinical trial.

Phase 2 is the most crucial part of the drug discovery process because the success rate is often quite low.

“Around 66 percent of trials fail in phase 2, where you measure efficacy of a drug in humans. So we use inClinco to predict which medicines will be successful in passing phase 2, thus increasing the efficiency of the development process and saving a lot of money on resources and time on the development,” Zhavoronkov explained.

“If a drug fails in phase 2, it is essentially a wasted effort and nothing comes out of it. We apply Inclinico to increase the likelihood of success for any programme we put forward.”

The US’ Food and Drug Administration (FDA) approved only 50 drugs last year, according to Zhavronkov, who is also a highly acclaimed scientist in the field of generative chemistry and biology. Only seven of those 50 drugs are “more or less innovative” as the rest are just repurposed meds, he added.

Because drug discovery is such a lengthy and costly process with such a huge probability of failure, no drugs have been discovered in the Middle East so far have been approved in the US. Several Big Pharma companies’ drugs rarely make it to phase 1 clinical trials.

However, with the power of AI, Insilico was able to nominate nine drugs last year alone, several of which made it to phase 1 clinical trials – a major difference when compared to Big Pharma companies which nominate four or five on average each year.

“We basically doubled the average of a big pharmaceutical company,” Zhavronkov said. “Double the productivity of a Big Pharma company on a tiny budget.”

No AI-designed drugs have ever been approved by the FDA yet because many of those efforts started around 2019 and 2020, although there were already some early players in this market before then.

Insilico started using generative AI in 2016 and it took them around five years to validate the drug discovery technology to ensure it can be trusted.

“As much as ChatGPT does a very good job of providing you with snappy responses to pretty much any question, many of those answers cannot be trusted,” Zhavoronkov said.

“In the pharmaceutical industry, you need to ensure that you have molecular precision and to achieve molecular precision, you need to have something like this, so we have multiple generative engines like ChatGPT that are chatting in the molecular language and providing you with desired properties. And then we need to evaluate those molecules and either reward or punish each generative model.”

The company began its own AI drug discovery journey in 2019 and its most mature project, a novel target molecule of AI for multiple types of fibrosis, just completed phase 1 and is going into phase 2. Insilico now has several programmes which tackle oncology, immunology, and the neural system, although none of them have made it to a clinic trial yet, it is important to note that only around four companies in the world have managed to reach human clinical trials with AI.

On average, the company spends about 72 hours to identify around 15 molecules, then synthesise, and eventually, test.

Zhavornokov published the first paper in generative AI-powered drug discovery in 2016. In the paper, he demonstrated for the first time that in theory, molecular structures in oncology can be generated. Later on, he published multiple papers in various journals.

However, AI-powered drug discovery has been met with scepticism from the scientific community. But in 2019, he released a paper on the topic in the journal Nature Biotech and made it onto MIT Technology Reviews’ top 10 breakthroughs of 2020.

“We can now create a pipeline of over 30 therapeutic programmes, 80 percent of them on cancer – at least 23 in cancer. Some are already in human clinical trials, so we actually have one oncology drug that just entered phase 1.”

“The time it takes to train the deep neural network, assemble the datasets, and design the algorithm is not comparable to the time it takes to validate. Because to validate just one feature, sometimes you might spend half a year and a few million dollars.

“Drug discovery just has longer validation cycles. So we have already made a revolution here.”

Middle East a ‘geographic priority’

“When Russia invaded Ukraine, we moved a lot of people from both regions into Abu Dhabi. We got really good support from the local government and decided to make it our base for AI. We are still hiring some of the top people in quantum computing and AI from Easter Europe, we call them kind of ‘AI refugees’.”

“Currently, we have about close to 50. BY the end of the year, we will probably be close to 75 and we’re scaling up so this region is extremely important for us for many reasons.”

Zhavornokov said that the Middle East is a “geographic priority” for Insilico, which is why it set up a base in Abu Dhabi. Another reason was because it has plans to train local talent and deploy its robotics lab in the UAE

Its robotics lab in China is the first of its kind and the only one of its kind in the world today that looks like something out of ‘Star Wars,’ the CEO said, and has six rooms fully automated with no need for human intervention.

“We started generating synthetic biological data using AI. Think of deepfakes but on our genetic, epigenetic and transcriptomic data so we have billions of people virtually simulated.”

Insilico intends to bring its algorithms to the UAE and retrain some of the generative biology tools.

“We’re launching a few this summer. It’s going to be everywhere, all over the news once we launch. But we actually want to focus on the Middle East because we are very serious about being based there. I don’t think there is any other AI-powered drug discovery company that ever even touched that region. But we decided to make it our own.”

The company also shared that it plans to launch a lab in the UAE in the next couple of years, similar to the fully AI-operated one it launched in China where the AI constantly trains itself to work with the utmost precision and accuracy.

“I promised this in China and in a year, I delivered it. We might be able to do this in the UAE, maybe not in a year but a couple of years because it will be much slower based on what we will need to be able to create a lab like that,” Zhavornokov said.

“This lab can be miniaturised, and we can put it in hospitals for personalised drug discovery. That’s my dream.”