Facebook unveils AI model to mix up cancer-curing cocktails with existing drugs

No longer content to simply provide a platform for keeping up with long-lost cousins and spreading conspiracy theories, Facebook has branched out into developing artificial intelligence to help treat complex diseases.

The social media giant’s AI research department and the Helmholtz Zentrum München, a research center in Germany focused on environmental health, unveiled an open-source AI model designed to determine the viability of repurposing existing drugs into new pharmaceutical cocktails.

Researchers and biologists now have free access to the Compositional Perturbation Autoencoder, or CPA, which evaluates the effects of drug combinations in varying dosages—a complicated task, as the number of possibilities can accelerate exponentially into the billions as more medicines are thrown into the mix.

The model predicts not only how the drugs interact with one another, but also how they might work together to attack specific cell types and interrupt diseases.

RELATED: Northwestern publishes open AI that spots COVID-19 in chest X-rays

The researchers trained the machine learning model on single-cell RNA sequencing data, to help it gauge the effects of drug cocktails on individual cells without requiring drug- or cell-specific programming.

Once trained, the model was tested on five RNA sequence datasets focused on cancer cells. The researchers found that CPA’s predictions of how drug combinations and dosages would affect those cells “reliably matched those found in the testing data set,” according to a Facebook AI blog post.

CPA can automatically evaluate dozens of drugs in various combinations and dosages in just a few hours, in stark contrast to the many years and repeated manual experimentation it would take to complete the task without machine learning.

“Our field has been successful in putting together cell atlases for different organs. Personally, I am now excited about leveraging those to predict tissue and cell-type-specific impact of drugs as well as their combinations,” Fabian Theis, part of the team behind CPA and head of Helmholtz’s Institute of Computational Biology, said in the blog post.

“This search space—across cell types, drug combinations as well as patient variation—is incredibly large and can never be explored in full experimentally, so machine learning is crucially needed here, and we contribute to this important step with our most recent approach,” Theis said.

RELATED: New approaches to AI, cell biology may have uncovered a novel drug target for COVID-19: Humans themselves

Though Facebook and Helmholtz claim that theirs is the “first single AI model” to predict the effects of drug combinations and dosages, several other AI developers have in fact produced similar models.

In just one example, last year, Notable Labs launched a long-term clinical trial to validate its own precision oncology platform, which uses AI to determine the drug or drugs that might be most effective to treat each cancer patient.