Profluent is trying to move AI-designed proteins from research papers into pharmaceutical work. The company, launched by ProGen researcher Ali Madani, is building models meant to design gene-editing and protein-producing systems for future medicines.
From Salesforce research to a new company
Salesforce, best known for cloud sales support software and Slack, previously spearheaded ProGen, a project focused on using generative AI to design proteins. Researchers behind the effort said in a January 2023 blog post that the approach could help uncover medical treatments more cost effectively than traditional methods if it reached the market.
ProGen produced research published in Nature Biotech showing that AI could create the 3D structures of artificial proteins. But after that research, the project did not become a commercial product at Salesforce or elsewhere.
That gap is what Profluent is now trying to address. Madani describes the company’s mission as “reversing the drug development paradigm,” beginning with patient and therapeutic needs and working backward toward “custom-fit” treatment solutions.
The idea is not to treat AI as a side tool for biology, but as a way to generate biological systems that are designed for a specific purpose. In Madani’s words, “Many drugs — enzymes and antibodies, for example — consist of proteins.” He added, “So ultimately this is for patients who would receive an AI-designed protein as medicine.”
Why proteins are a fit for generative AI
The scientific logic behind Profluent comes from a comparison Madani explored while working at Salesforce’s research division. Natural language, such as English, has patterns that AI systems can learn. Proteins also have patterns.
Proteins are chains of bonded-together amino acids. The body uses them in many ways, including making hormones and repairing bone and muscle tissue. Madani found that protein information could be approached somewhat like words in a paragraph.
When data about proteins is fed into a generative AI model, the model can be used to predict entirely new proteins with novel functions. For drug discovery, that matters because the desired medicine may not already exist in nature in the exact form researchers need.
Other groups have also shown that generative AI can be applied to proteins and molecules. Nvidia released MegaMolBART in 2022, a generative AI model trained on millions of molecules to search for potential drug targets and forecast chemical reactions. Meta trained ESM-2 on protein sequences and said the approach let it predict sequences for more than 600 million proteins in just two weeks. DeepMind, Google’s AI research lab, has AlphaFold, which predicts complete protein structures with speed and accuracy beyond older, less complex algorithmic methods.
The gene-editing focus
Profluent is extending the protein-design concept into gene editing. Madani co-founded the company with Alexander Meeske, an assistant professor of microbiology at the University of Washington.
The company’s case is that many genetic diseases cannot be addressed simply by taking existing proteins or enzymes from nature. Madani also argues that gene editing systems assembled by mixing and matching capabilities can face tradeoffs that limit how broadly they can be used.
Profluent’s answer is to design gene editors around multiple desired attributes at the same time. Madani said the company can optimize those attributes “to achieve a custom-designed [gene] editor that’s a perfect fit for each patient.”
The startup is training AI models on data sets with over 40 billion protein sequences. Those models are intended to create new gene-editing and protein-producing systems, as well as fine-tune existing ones.
Profluent does not plan to develop treatments entirely on its own. Instead, it intends to work with outside partners to produce “genetic medicines” that have the most promising paths to approval.
What could change for drug development
The company’s broader promise is about reducing friction in the early stages of medicine creation. Madani argues that Profluent’s approach could cut down the time and capital usually required to develop a treatment.
According to industry group PhRMA, it takes 10-15 years on average to develop one new medicine from initial discovery through regulatory approval. Recent estimates put the cost of developing a new drug between several hundred million and $2.8 billion.
Those figures explain why AI drug discovery has drawn interest. If models can produce better starting points for medicines, companies may be able to spend less time searching and more time testing the most promising candidates.
Madani frames the shift as a move away from chance. “Many impactful medicines were in fact accidentally discovered, rather than intentionally designed,” he said. He added that Profluent’s capability gives “humanity a chance to move from accidental discovery to intentional design of our most needed solutions in biology.”
Funding, partners and pressure from rivals
Profluent is based in Berkeley and has 20 employees. Its backers include Spark Capital, which led the company’s recent $35 million funding round, along with Insight Partners, Air Street Capital, AIX Ventures and Convergent Ventures.
Google chief scientist Jeff Dean has also contributed, adding credibility to the company’s platform. The startup’s near-term focus is upgrading its AI models, including by expanding training data sets, as well as acquiring customers and partners.
Profluent will not have the field to itself. EvolutionaryScale and Basecamp Research are also training protein-generating models and raising large amounts of venture capital.
Madani says the company has already built its initial platform and shown scientific breakthroughs in gene editing. The next phase is about scale: “Now is the time to scale and start enabling solutions with partners that match our ambitions for the future.”