$8,9 billion for AI drug development, while evidence infrastructure was left for later

The path from an AI recommendation to an FDA submission must be documented from the first experiment, writes Moe Alsumidaie

On August 8, Moe Alsumidaie, editor in chief of the industry publication The Clinical Trial Vanguard, published a column about the gap between early AI driven molecule discovery and the preparation of a drug application. Citing Dealforma, he reports that companies developing drugs with AI raised $8,9 billion across 264 funding rounds in 2024. His subject is the path from a software recommendation to the documents submitted to the FDA.

At an early stage, AI helps researchers select the target that a future drug should act on, identify candidate molecules, and predict their properties. The team then selects a molecule for laboratory validation and subsequent testing in humans. Alsumidaie argues that the submission package must preserve the full record of this decision: which biological data were used, which model version produced the recommendation, which experiment confirmed it, and who decided to continue development.

This documentation links the future submission to a specific experiment, and the experiment to the data from which the model's recommendation emerged. Data become evidence when they can be traced back to the cell line, sample, instrument, and protocol. The current column extends this chain: a validated sample becomes the basis for a decision about the molecule.

Alsumidaie writes in his column:

“The question worth watching is not which AI platform will secure the first drug approval, but which developer will build the data infrastructure needed to prove its case,” Alsumidaie writes.

On January 14, the FDA and the European Medicines Agency issued ten shared principles for the use of AI in drug development. The document recommends recording the origin and processing of data, decisions made during analysis, the intended purpose and conditions of AI use, and updates to the system itself throughout the drug's development.

In Alsumidaie's view, the advantage will go to the developer that maintains this chain from the first experimental record and can submit it with the application. The first approval will show the path taken by one candidate.