Image credit: koto_feja / Getty Images
Last year, Salesforce, the company best known for its cloud sales support software (and Slack), spearheaded a project called ProGen to design proteins using generative AI. A research moonshot, ProGen — if brought to market — could help make medical treatments more effective than conventional methods, the researchers behind it claimed in a January 2023 blog post.
ProGen’s research, published in the journal Nature Biotech, demonstrated that AI can successfully generate 3D structures of synthetic proteins. But, off paper, the project wasn’t much at Salesforce or anywhere else — at least not in a commercial sense.
That is, until recently.
Ali Madani, one of the researchers responsible for ProGen, has started a company, Profluent, that he hopes will bring similar protein-generating technology out of the lab and into the hands of pharmaceutical companies. In an interview with TechCrunch, Madani described Profluent’s mission as “changing the paradigm of drug development,” starting with patient and therapeutic needs and building “custom-fit” therapeutic solutions. Working backwards.
“Many drugs — enzymes and antibodies, for example — are made up of proteins,” Madani said. “So ultimately it’s for patients who will receive AI-designed proteins as medicine.”
While in the research division of Salesforce, Madani found himself drawing parallels between natural language (e.g. English) and the “language” of proteins. Proteins—chains of amino acids linked together that the body uses for a variety of purposes, from making hormones to repairing bone and muscle tissue—can be thought of like words in a paragraph, Madani discovered. . Fed into a generative AI model, data about proteins can be used to predict entirely new proteins with new functions.
Perfluent, along with Madani and co-founder Alexander Mesik, an assistant professor of microbiology at the University of Washington, want to take this concept a step further by applying it to gene editing.
“Many genetic diseases cannot be cured [proteins or enzymes] picked up directly from nature,” Madani said. “Furthermore, mixed and matched gene editing systems for new capabilities suffer from functional trade-offs that significantly limit their accessibility. In contrast, Profluent can optimize multiple attributes simultaneously to achieve a custom design. [gene] An editor tailored to each patient.
It’s not out of left field. Other companies and research groups have demonstrated viable ways in which creative AI can be used to make protein predictions.
Nvidia released MegaMolBART, a creative AI model in 2022, that was trained on data sets of millions of molecules to find potential drug targets and predict chemical reactions. Meta trained a model called ESM-2 on protein sequences, which the company claimed allowed it to predict the sequences of more than 600 million proteins in just two weeks. And DeepMind, Google’s AI research lab, has a system called AlphaFold that predicts complete protein structures, achieving speed and accuracy far greater than older, less complex algorithmic methods.
Profluent is training AI models on massive datasets—datasets with more than 40 billion protein sequences—to build new as well as fine-tune existing gene-editing and protein-generating systems. Instead of developing the treatment itself, the startup plans to collaborate with external partners to get “genetic drugs” with the most promising pathways to approval.
Madani claims this approach can dramatically reduce the amount of time and capital typically required for treatment. According to industry group PhRMA, it takes an average of 10-15 years to develop a new drug from initial discovery through regulatory approval. According to recent estimates, the cost of developing a new drug ranges from several hundred million to 2.8 billion dollars.
“Many effective drugs were actually discovered by accident rather than by deliberate design,” Madani said. “[Profluent’s] Competence offers humanity an opportunity to move from accidental discovery to the intentional design of our most essential solutions in biology.
Based in Berkeley, the 20-employee Perfluent is backed by VC heavy hitters including Spark Capital (which led the company’s recent $35 million funding round), Insight Partners, Air Street Capital, AIX Ventures and Convergent Ventures. Jeff Dean, Google’s chief scientist, has also contributed, lending additional credibility to the platform.
Madani says Profluent’s focus in the next few months will be on upgrading its AI models, in part by expanding training datasets, and customer and partner acquisition. He will have to move aggressively. Competitors, including EvolutionaryScale and Basecamp Research, are rapidly training their protein-generating models and raising large sums of VC cash.
“We have developed our initial platform and shown scientific breakthroughs in gene editing,” Madani said. “Now is the time to start scaling and enabling solutions with partners aligned with our ambitions for the future.”