AI has yet to fulfill its promise of helping treat or even eradicate some of the most serious diseases. Two initiatives announced this week are hoping to change that.
Less than five months after its launch, French startup Bioptimus announced the release of H-optimus-, which is being billed as the world’s largest open-source AI foundation model for pathology. With 1.1 billion parameters, H-optimus-0 is trained on a proprietary dataset of several hundreds of millions of images extracted from over 500,000 histopathology slides across 4,000 clinical practices. This sets a new benchmark for state-of-the-art performance in several critical medical diagnostic tasks, from identifying cancerous cells to detecting genetic abnormalities in the tumor.
Meanwhile, Cambridge, Massachusetts-based Flagship Pioneering, the venture capital fund that launched Moderna,an American pharmaceutical and biotechnology company that focuses on RNA therapeutics, primarily mRNA vaccines , said it is planning to create multiple new biotech companies harnessing GenAI to power drug discovery after securing $3.6 billion in funding from its financial backers.
Traditional drug discovery is a notoriously time consuming and expensive process, with pre-clinical stages typically taking three to six years and costing hundreds of millions to billions of dollars. With the promise of lower costs and shorter development timelines, AI-enabled drug discovery holds massive potential to increase the accessibility of drugs and to treat presently incurable conditions.
AI’s application to pathology, the cornerstone of disease diagnosis which requires meticulous examination of tissue samples to identify abnormalities, also appears promising. Traditionally, this process relies heavily on pathologists’ expertise and experience. However, the increasing complexity and volume of cases necessitate advanced tools that can assist pathologists in making faster, more accurate diagnoses.
“H-optimus-0 is just the beginning,” Professor Jean-Philippe Vert, PhD, co-founder and CEO of Bioptimus, said in a statement. “It marks the first in a long series of models we will create at Bioptimus, each more advanced and comprehensive than the last. Future models will not only be trained on an even larger number of pathology images from Europe, Asia and Africa but will also incorporate other modalities, such as genomics and proteomics. Our ultimate goal is to create the first multiscale foundation model of biology, capable of integrating diverse biological data and scales to enable scientific discoveries and accelerate biomedical innovations.”
While a few startups are leveraging LLMs for specific areas of biology, such as the design of novel protein therapeutics, Bioptimus believe it is the first to try and build a model that will be trained on data that’s necessary to understand multiple biological processes and how they connect with each other. The aim is to address many different scales of biology, including organs, tissues, cells, molecules, and atoms, to gain a holistic view of how the human body functions and advance the treatment of disease. In short, it wants to use LLMs to try and unlock the language of life.
Even simple biological systems are made up of a huge number of components that interact with one another in complicated ways that are not yet understood. The hope is that amassing of huge datasets and the application of AI will allow researchers to gain fuller, more accurate pictures of complex biological systems.
One of the key forces behind the dramatic recent progress in artificial intelligence is so-called “scaling laws” : the fact that radical improvements in performance result from continued increases in LLM parameter count, training data and compute.
In the short-term applying LLMs to patient data could help accelerate the development of new drugs and precision medicine, says Vert. Longer term it has the potential to help create digital twins of individuals to capture and monitor the state of the body, making disease prevention easier, he says.
To achieve those goals Bioptimus is assembling a team of scientists that includes Google DeepMind alumni as well as scientists from Owkin, a French unicorn and a member of the World Economic Forum’s Global Innovators Community.
DeepMind, which is owned by Google parent company Alphabet, in late 2020 used an AI system called AlphaFold (which does not use large language models) to produce a solution to protein folding that humans could not solve. Because a protein’s shape is closely linked with its function, knowing a protein’s structure unlocks a greater understanding of what it does and how it works, helping accelerate scientific research and discovery globally. Within 12 months AlphaFold was accessed by more than half a million researchers and used to accelerate progress on important real-world problems ranging from plastic pollution to antibiotic resistance.
Owkin, for its part has been using early versions of AI to identify new treatments, de-risk and accelerate clinical trials and develop AI diagnostics. It founded MOSAIC, the world’s largest multi-omics atlas for cancer research. Multi-omics is a new approach that combines the data sets of different modalities: in this case bulk, single-cell and spatial transcriptome, genome, pathology slide, and clinical data.
Alumni from both companies who are now working for Bioptimus are optimistic that they can make further – potentially radical – improvements to healthcare by applying LLMs to biology.
Accelerating Drug Development
Much of the investments from Flagship’s fund VIII will be focused on platforms that use AI in novel ways to develop drugs, Noubar Afeyan, who founded Flagship in 2000, told The Financial Times.
Ideas that Flagship are developing include using AI to generate novel chemical materials which serve as the building blocks of medicines, as well as using the technology to automate hypothesis generation for scientists. “AI is a prosthesis for your imagination — your ability to leap dramatically changes using these tools,” said Afeyan. The money from Flagship’s eighth fund will be spent over the next three years.
In the Convention on Pharmaceutical Ingredients (CPHI) Annual Report 2023, which provides insight from 250 global pharmaceutical companies, some 60% of executives identified AI drug discovery as a technology that will be ‘used routinely in 2026’, with 42% forecasting that the first ‘FDA-approved drug discovered by AI’ will be seen in the next 2-5 years. About 20% believe it will happen even sooner.
Looking further ahead, by 2030, some 52% of new drugs approved are expected be discovered or developed using AI.
The report says pharmaceutical ‘AI companies’ have overtaken ‘late stage’ and ‘early stage’ biotechs as the industry’s most appealing investment option for venture capital.
IN OTHER NEWS THIS WEEK
QUANTUM COMPUTING
EDF Partners With French Research Institute And Startups To Optimize Quantum Computing’s Energy Consumption
The French electric utility company EDF, along with quantum computing companies Quandela and Alice & Bob, and the French National Centre for Scientific Research (CNRS),announced July 10 that they have signed an agreement to collaborate on a project to optimize energy consumption in quantum computing.
The project, named “Energetic Optimization of Quantum Circuits” (OECQ), will, in the first phase, compare the energy requirements of high-performance computing (HPC) systems with those of quantum computers. This analysis will be based on relevant industry use cases and the advanced computations they require. This part of the project will provide the first measurement of energy consumption for the full stack of a quantum computer.
In the second phase, the project will focus on optimizing the energy consumption of quantum computers, addressing not only the energy required by the quantum processing unit (QPU) per se but also that of the auxiliary technologies powering the QPU, which are expected to make up a relevant share of the total quantum system consumption.
The OECQ project gathers a cross-disciplinary team, pooling together fundamental research and industry R&D:
Energy company EDF offers industry use cases, as well as classical and quantum computing expertise. EDF will identify the relevant use cases that demand intensive computing power and therefore potentially high energy consumption.
Alice & Bob specializes in quantum processing unit and algorithm development. They will test the energy needs of their innovative superconducting qubit architecture, the cat qubit.
Quandela is a leader in photonic quantum computing providing industry-grade hardware and software solutions. Their platform is based on semiconductor qubit devices and integrated photonics.
Alice & Bob and Quandela will estimate the energy consumption that a quantum algorithm would require for the selected use cases if solved using their current quantum systems. They will then leverage the insights gained, to build and test new, more energy-efficient quantum processor prototypes.
The CNRS Quantum Energy Team provided the first methodology to tackle energy costs of the full stack of a quantum computer.
The project, with a total cost of €6.1 million, is supported up to €4.5 million by a France 2030 grant operated on behalf of the French state by Bpifrance, France’s public investment bank.
“One of the primary objectives of the OECQ project is to seize the opportunity to develop quantum computing technology in an energy-efficient manner from the outset,” Théau Peronnin, CEO of Alice & Bob said in a statement. “As quantum computing is still in its nascent stages, this project will dimension the future energy infrastructures that will support mature quantum technology. Additionally, since energy consumption is a key cost driver in quantum tech, making these processes more efficient offers a significant competitive advantage.”
Reversible Computing Could Drastically Reduce Computing’s Energy Needs
One of the requirements for efficient quantum computers is reversible computing. In the reversible computing model, operations that process logic can be reversed and recapture energy. Earlier this month Vaire Computing, a reversible computing company based in London and Seattle, announced that it had raised $4 million in a seed round to work on building silicon chips that would consume negligible amounts of energy and generate little heat, if any.
ARTIFICIAL INTELLIGENCE
AMD To Acquire Europe’s Largest Private AI Lab
Advanced Micro Devices 9AMD) said July 10 it will acquire Finland’s Silo AI, Europe’s largest AI lab, for about $665 million in cash as the company tries to enhance its AI chip capabilities to compete against industry leader Nvidia. Acquiring Silo AI will help AMD improve the development and deployment of AMD-powered AI models and help potential customers build complex AI models with the company’s chips, AMD said. Silo AI specializes in end-to-end AI-driven solutions that help customers integrate the tech into their products and services. With operations in Europe and North America, the startup counts Philips, Rolls-Royce and Unilever among its customers. “Across every industry, enterprises are looking for fast and effective ways to develop and deploy AI solutions for their unique business needs,” Vamsi Boppana, senior vice president of the Artificial Intelligence Group at AMD, said in a statement. “Silo AI’s team of trusted AI experts and proven experience developing leadership AI models and solutions, including state-of-the-art LLMs built on AMD platforms, will further accelerate our AI strategy and advance the build-out and rapid implementation of AI solutions for our global customers.”
To access more of The Innovator’s News In Context stories click here.
