The New Interface of Gen AI for Drug Discovery

Let's face it, 2023 is a year a lot of us will remember as a technological tsunamy driven by ChatGPT public launch. And lots and lots of other models that followed. But ChatGPT did the crucial heavylifting: for the first time in history of artificial intelligence (AI) field, general non-specialist public suddenly realized AI is a real thing, and you don't need to have PhD in Computer Science to use it. ChatGPT is not perfect, with limitations, errors, hallucinations, safety concerns... but it is a big deal -- if you do basic training in prompt engineering and read a dozen relevant articles to figure out what you can do and how.

What is coming in 2024 is even more interesting. Some critics say, large language models are destined to fall or at least experience market correction, because of high cost of training models and vague prospects for earning stable revenue. Others say, LLMs will reshape the world, all our digital habits will be completely upside down, and even Google search might disappear eventually.

For instance, according to Dr John Bates—a pioneer in the fields of the Internet of Things and Big Data Streaming Analytics event-driven architectures, smart environments and real-time computing—we may see the rise of "sentient documents", the "death of email", and bots as proxies at meetings. Who knows. I love emails for their stability, I hope they don't fall.

Anyway, back to today's topic: generative AI in drug discovery.

Recently, I visited Pharma.AI day by Insilico Medicine, a Hong Kong/New York-based pioneer of generative AI for drug discovery.

Pharma.AI is an end-to-end drug discovery platform that includes three key modules: PandaOmics for building hypotheses and target discovery; Chemistry42 for generative drug design; and Medicine42 for predicting success and optimizing clinical trials.

This platform is now used by over 40 leading pharma companies, and has produced an internal pipeline of 31 therapeutic assets across 29 targets, including in cancer, fibrosis, and central nervous system diseases, with four in clinical stages. The Company’s lead drug, designed to treat idiopathic pulmonary fibrosis, is the first AI-designed drug for an AI-discovered target to reach Phase II trials with patients. Another drug developed by this platform – a potentially best-in-class USP1 cancer inhibitor for BRCA-mutated tumors – was recently licensed to Exelixis for $80m upfront and additional milestone and royalty payments.?

New features of their generative AI platform had "wow" effect on me. But first things first.

There was a “meaty” presentation by Alex Zhavoronkov founder and CEO of Insilico, and I just printscreened a couple of points from his slides I guess worth mentioning:

For instance, with Insilico’s Pharma.AI, it is possible to do this:

Alex also pointed out that in the AI era, those who can produce proprietary biological and chemical data at scale will gain a strategic advantage. Insilico has built a fully automated robotic laboratory for biology research and drug design in the city of Suzhou, China.

The lab features many things, including sample registration, cell cultures, compound management, HTS, high content imaging, NGS, etc.

Appartently, they care about the future of their AI business model.

Now, some of the new features added to Pharma.AI suite:

Randomly what I remember:

There is now an indication prioritization tool. Also, there is a new feature of automated gene-disease reports where you can crawl and summarize the association data with Gen AI.

In Chemistry42 there is now a new interactive way to set up experiments in the laboratory. Also, there are now 26 new models for ADMET predictions (generation, optimization, annotation). There are new UI components to create new reward functions.

There is new Alchemistry system, it calculates free binding energy (BE) for the generated molecules, compare molecules by BE. In theory don’t need any additional physics tools.?

Also, there is new module for kinases:

and lots of updates to inClinico software:

But the one feature that is especially cool is Copilot, which is essentially ChatGPT for drug discovery.

The Copilot is connected to Pharma.AI platform and is fine-tuned on Insilico’s data. What it means is that you can basically talk to the platform and search for new ideas that you can then explore further. With this, you do not even need to know how it works. You just need to be a good biologist and know what to ask and where to move the conversation. The system has all the data and lots of potential answers.

I am a big fan of generative AI, and I am using dozens of tools myself. We also integrated custom GPTs into our BiopharmaTrend.com market research platform. And I do believe this is the future of interaction with digital resources and databases.

So, I liked this new feature a lot. It is nice when you can have a meaningful conversation with your lab software!

---

Welcome to my newsletter, "Where Technology Meets Biology." I am sharing noteworthy news, trends, biotech startup picks, industry analyses, and interviews with pharma KOLs. Contact me for consulting or sponsorship opportunities here or at www.BiopharmaTrend.com.

Enjoying the newsletter? Subscribe to become part of 11K+ readers here on LinkedIn. Please help us spread the word by sharing it with your colleagues and friends.

Also, consider joining my Substack community where we are exploring a lot more (4.4K+ industry professionals are eading it via email).

-- Andrii