Atomic launches to apply AI to small molecule drug discovery
Atomic AI raised $35m (€32m) from a Series A financing round led by Playground Global. The capital adds to a previous $7m seeding round led by 8VC, adding up to $42m in funding to pursue its development.
At the heart of the biotech, and the reason for it generating funds, is its proprietary 3D RNA structure engine that generates RNA structural datasets. Once machine learning is applied, alongside in-house experimental wet-lab biology, Atomic is enabled to carry out RNA drug discovery.
The biotech’s platform can discover structure, ligandable RNA motifs, which it states has proven to be a critical barrier to current approaches in small molecule drug discovery against RNA. Atomic added that it also has the capacity to design RNA-based medicines and tools.
According to the company, by using RNA foundation models that operate directly at the atomic level with custom data, the platform can discover novels structural targets across transcripts of interest. This enables the drugging of the transcriptome, and to generate new structural motifs for use in RNA medicines and tools.
Gene Yeo, professor of Cellular and Molecular Medicine at the University of California San Diego, said: “There are well-known RNA sequences that play a critical role in driving disease; however, current technologies lack the RNA-structure modeling capabilities necessary to create confidence in RNA-targeting drug discovery.
"Atomic has created a sophisticated and integrated AI engine that will transform the discovery and development of RNA-targeted medicines.”
3D RNA structures
The biotech outlined that it plans to use its database of discovered and designed 3D RNA structures to develop a pipeline of small molecule drug candidates.
The research behind the company emerged from the work of its CEO, Raphael Townshend, and others within the biotech, completed at Stanford University.
During its early stages, those working on the platform revealed that the team allowed the machine learning technology to understand what makes structural predictions accurate more or less accurate by itself. This allowed the algorithm to discover features that may be missed if researchers picked out the specific features themselves.
“Most of the dramatic recent advances in machine learning have required a tremendous amount of data for training. The fact that this method succeeds given very little training data suggests that related methods could address unsolved problems in many fields where data is scarce,” said Ron Dror, associate professor of computer science at Stanford University, and advisor to the biotech.