Accelerating Scientific Research

One bottleneck after a large gene experiment is working out what the results mean. In an Anthropic case study, Iain Cheeseman's lab at the Whitehead Institute and MIT uses CRISPR gene editing to knock out thousands of genes, photographs the affected cells, and uses software to group genes whose removal produces similar-looking changes. Interpreting each group, meaning what biological process its genes might share, normally takes an expert combing the literature, and most groups are never investigated. The lab built a Claude-powered system called MozzareLLM to take over part of that step. It proposes a shared process, flags which genes are well understood and which are poorly studied, and highlights candidates for follow-up, with confidence levels that Cheeseman says help decide whether further resources are worth spending. The company's account is an attributed description of one lab's tool. Its claims that Claude outperformed other models and catches things the lab missed are reported without published data here.