Converge Bio, a Boston- and Tel Aviv-based startup leveraging generative AI to accelerate drug discovery, has raised $25 million in a Series A round led by Bessemer Venture Partners. The oversubscribed funding round also included participation from TLV Partners, Vintage Investment Partners, and undisclosed executives from Meta, OpenAI, and Wiz. The company focuses on helping pharmaceutical and biotech firms shorten drug development timelines by integrating AI into key stages of the R&D process. Converge trains generative models on biological data such as DNA, RNA, and protein sequences, then embeds these models into clients’ workflows to generate and optimize potential drug candidates. CEO and co-founder Dov Gertz emphasized that the company’s platform is designed to be seamlessly integrated, not a collection of isolated models. “Our systems are built as cohesive pipelines,” he said. “For example, our antibody design system combines a generative model for creating novel antibodies, predictive models to assess molecular properties, and a physics-based docking system to simulate how the antibody interacts with its target.” Converge has already launched three customer-facing systems: one for antibody design, another for improving protein yield, and a third for identifying biomarkers and drug targets. The platform’s strength lies in its end-to-end integration, allowing partners to skip the complexity of building and connecting models themselves. Since its founding two years ago, Converge has grown rapidly. It has signed 40 partnerships with pharma and biotech companies, currently running about 40 active programs. The team has expanded from nine to 34 employees, and the company is now extending its reach into Asia, in addition to its existing operations in the U.S., Canada, Europe, and Israel. The startup has also published public case studies demonstrating its impact. In one, a partner achieved a 4 to 4.5 times increase in protein yield in a single computational run. In another, Converge generated antibodies with binding affinities in the single-nanomolar range—indicating high effectiveness. Gertz noted a dramatic shift in industry sentiment. When the company launched, skepticism was common. Now, the momentum is undeniable, driven by high-profile successes like the Nobel Prize awarded to AlphaFold’s developers in 2024 and major investments such as Eli Lilly’s collaboration with Nvidia to build a dedicated AI supercomputer for drug discovery. Despite the excitement, challenges remain. Large language models can produce inaccurate or biologically implausible molecules—what Gertz calls “hallucinations.” These are harder to detect in molecular design than in text, where errors are more obvious. To mitigate this, Converge combines generative models with predictive and physics-based systems to filter and validate candidates, significantly reducing risk. Gertz acknowledged concerns raised by AI pioneers like Yann LeCun, who question the reliability of LLMs in scientific domains. “I respect LeCun’s views,” he said. “We don’t rely on text-based models for core scientific insight. Our models are trained on biological sequences and molecular data, not text.” Text-based LLMs are used only as supplementary tools—for example, to help researchers explore scientific literature on generated compounds. The company uses a flexible approach, incorporating LLMs, diffusion models, traditional machine learning, and statistical methods depending on the task. “We’re not tied to one architecture,” Gertz said. Looking ahead, Gertz envisions a future where every life sciences organization uses Converge Bio as its generative AI lab. “Wet labs will remain essential, but they’ll be paired with computational generative labs that create hypotheses and molecules in silico,” he said. “Our goal is to become that foundational lab for the entire industry.”