The Virtual Biotech releases an inspectable framework for research agents

Its September 17 Science paper describes three uses: retrospective trial analysis, a proposed lung-cancer strategy and investigation of a terminated ulcerative-colitis trial. [1]

In the retrospective analysis, more than 37,000 agents annotated outcomes from 55,984 trials. The large count describes the extent of the analysis; it does not mean that thousands of new treatments were tested. The other examples concern generating a strategy and inferring possible reasons for failure. They show the kinds of research question the framework addresses, while leaving prospective therapeutic efficacy to later work. [1]

The public implementation makes the approach inspectable. Its documentation, rechecked on September 28, recommends a conversational interface and supplies setup and diagnostic checks. The core installation calls for about forty gigabytes of storage and recommends sixteen gigabytes of memory. Ordinary use relies on model APIs rather than requiring a local GPU, and research queries incur API charges. Code availability therefore reduces one barrier without making the computation or reference-data preparation costless. [3]

For a first evaluation, ask it to reconstruct an answer whose evidence is already well documented. Check whether each claim leads to the relevant record, associations remain distinct from mechanisms, and negative findings survive the synthesis. These tests precede any claim of discovering a medicine.