While there are currently no FDA-approved AI-discovered drugs, interest in generative AI (genAI) could be a game-changer for the pharmaceutical industry in the coming years. Sanofi, which has signaled its intent to go “all in” on AI, has inked a deal with ChatGPT developer OpenAI and Formation Bio to build AI-powered software to speed drug development.
“We are always thinking about ways to drive innovation forward – ourselves and with partners,” Emmanuel Frenehard, EVP and chief digital officer, Sanofi told Drug Discovery & Development via email. “Together, we bring complementary resources and abilities to the table that enable us to jointly push ahead in AI innovation for drug development more efficiently than any one company could do alone.”
Why Sanofi, OpenAI, Formation Bio joined forces
Emmanuel Frenehard, EVP and Chief Digital Officer, Sanofi
Under the partnership, Sanofi will provide proprietary data to develop AI models, OpenAI will offer advanced AI capabilities and expertise, and Formation Bio will lend its engineering resources and its tech-driven development platform. The aim is to create custom AI agents (automated AI-enabled programs) and models tailored for pharma needs.
Paul Hudson, CEO of Sanofi, described the collaboration in a press release as “the next significant step in our journey to becoming a pharmaceutical company substantially powered by AI.”
In a LinkedIn post, Hudson noted that the alliance would “tap into the respective strengths of our three organizations to develop next-gen, customized Artificial Intelligence solutions aimed at improving productivity and accelerating the drug development lifecycle.” In particular, Hudson mentioned the fact that AI was “already an everyday reality for many of our scientists at Sanofi.” The new partnership aims to we “go even further by building new foundations for drug development overall.”
Meanwhile, Benjamine Liu, Co-Founder & CEO of Formation Bio, said in a press release the partnership will enable the three companies to “reimagine drug development.”
Similarly, Brad Lightcap, COO of OpenAI, noted the “massive potential for AI to accelerate drug development.” He expressed excitement about collaborating with Sanofi and Formation Bio “to help patients and their families by bringing new medicines to market.” OpenAI CEO has described the potential of AI to help “cure” all disease.
Open-source and proprietary genAI tools have also emerged recently
BioMistral: Several other initiatives have recently emerged in the life sciences space. BioMistral, a collection of open-source large language models trained on biomedical data from PubMed Central including several 7B parameter models, have shown strong performance in medical question answering and multilingual capabilities. A preprint introduces BioMistral 7B, “a specialized LLM tailored for the biomedical domain.”
SciBite Chat: Separately, SciBite, an Elsevier subsidiary, has debuted SciBite Chat, a genAI-powered search tool for life science researchers. By combining semantic search, genAI (including support for OpenAI’s GPT-4 models) along with structured data ontologies and curated vocabularies, SciBite Chat aims to streamline biomedical research while minimizing the risk of inaccurate outputs. “We combined the best of the LLM, the generative AI with our structured data,” said Jane Lomax, Ph.D., head of ontologies at Elsevier’s SciBite, in a recent interview. “You get the ability to ask a natural language question, but you get the explainability of being able to use that structured data behind the scenes. So you get the best of both worlds.”
Nach0: In addition, Insilico Medicine and NVIDIA have developed nach0, a new large language model (LLM) transformer that bridges the gap between biomedical and chemical tasks. Nach0 was trained on a diverse dataset, including PubMed abstracts, patent descriptions, and molecular structures, allowing it to perform natural language processing, chemistry-related tasks, and cross-domain tasks. Built on the NVIDIA BioNeMo generative AI platform, nach0 outperformed other LLMs in molecular tasks and significantly surpassed ChatGPT. Case studies demonstrated nach0’s ability to generate molecules for potential therapeutic activity and validate promising options for medicinal chemists.
Earlier, Insilico had launched Precious3GPT, a highly multimodal, multi-omics and multi-species transformer model developed for aging research, drug discovery, chronic disease research and biomarker development.
AlphaFold 3: Finally, Isomorphic Labs and DeepMind launched AlphaFold 3 earlier this month. The model can accurately model the structures of biomolecular complexes, including proteins, nucleic acids, DNA, RNA, small molecules, ions and modified residues, with a novel diffusion-based architecture that predicts raw atom coordinates.
Filed Under: clinical trials, Data science, Drug Discovery and Development, machine learning and AI
