Isomorphic Labs Inc., a healthcare startup that spun out of Alphabet Inc. in 2021, today announced that it has closed a $600 million funding round.

OpenAI backer Thrive Capital led the investment. It was joined by Alphabet and GV, the search giant’s startup investment arm.

Isomorphic Labs is led by Chief Executive Officer Demis Hassabis, who has led Google LLC’s DeepMind artificial intelligence lab. The company is building what it describes as an AI drug design engine. The software reduces the amount of manual work involved in developing new medicines and thereby speeds up the process.

Most drugs are based on so-called small molecules. Those are lightweight organic molecules that can disrupt a biological process associated with a disease. According to Isomorphic Labs, its AI streamlines the task of designing new small molecules with therapeutic applications.

The company’s software can also ease several other aspects of the drug development workflow. 

In addition to having therapeutic potential, a small molecule must meet many other criteria to effectively treat disease. One of those requirements is the ability to effectively attach to a disease-causing protein. In a blog post published last July, Isomorphic Labs executive Pascal Savy detailed that the company’s drug design engine can predict the effectiveness with which a small molecule attaches to a protein. It also maps out other properties such as a small molecule’s solubility.

Under the hood, the drug design engine is powered by several different AI models. One of them is Google LLC’s AlphaFold 3 model. It’s an improved version of AlphaFold 2, the AI system for which Isomorphic Labs CEO Demis Hassabis won half of last year’s Nobel Prize in chemistry with former DeepMind colleague John Jumper.

The behavior of proteins, the basic building blocks of life, is heavily influenced by their shape. Mapping out a single protein’s shape historically took several years. AlphaFold 2 reduced that process to a few minutes.

AlphaFold 3, the improved version of the model used by Isomorphic Labs’ drug design engine, debuted last May. It can predict the shape of not only proteins but also DNA, RNA and other biological building blocks. Moreover, the algorithm does so more reliably than earlier models.

The newer model is powered by a diffusion network, a type of neural network commonly used for image generation tasks. To predict the shape of a biomolecule, the model generates a digital object that represents a cloud of atoms. AlphaFold 3 then refines this atom cloud in multiple steps until it turns into a visualization of the molecule’s shape.

The AI system also incorporates an attention mechanism. This is a software module that large language models use to identify the most important items in a collection of data points. According to Google, the attention mechanism enables AlphaFold 3 to generate more accurate predictions about the shape of biomolecules than earlier technologies.

Isomorphic Labs is using its software to power several in-house drug development programs. In addition, the company has partnered with Eli Lilly an Co. and Novartis AG to support their research efforts. Last month, Isomorphic Labs announced plans to expand the collaboration with Novartis by adding up to three more joint research programs.

The Alphabet spinout will use the $600 million round announced today to enhance its software and finance drug development programs. Isomorphic Labs will hire more employees to support the effort. 

Image: Isomorphic Labs