Collaboration aims to reduce animal use in drug safety testing

Roughly a third of drug candidates fail in the first phase of clinical trials, often due to unanticipated toxic effects that weren’t apparent in animal tests. A consortium co-led by Broad Institute scientists is building tools and datasets to reduce the need for animal testing in drug and agricultural chemical development.

Launched in 2023 by Broad, the non-profit Health and Environmental Sciences Institute (HESI Global), and scientific partners spanning industry, government, academia, and other research institutes, and catalyzed by a grant from the Massachusetts Life Sciences Center, the Omics for Assessing Signatures for Integrated Safety (OASIS) consortium is using cell-based models and integrating cutting-edge cellular, transcriptomic, proteomic, and AI and machine learning technologies to better predict damaging effects of various compounds on the liver.

OASIS includes more than 50 partner groups and is two years into its initial three-year pilot phase. The researchers are testing compounds on multiple cell models, and are now producing and sharing datasets from their first rounds of experiments.

Consortium members are optimistic that their work will lead to tools and data resources that can make drug and agricultural chemical testing systems more accurate, efficient, and cost-effective, with the goal to reduce the reliance on laboratory animals while having more human-relevant data.

“Despite our reliance on them, animal tests often fail to accurately predict impact on humans,” said Shantanu Singh, an OASIS co-leader and senior group leader in the Imaging Platform at the Broad Institute. “Our goal now is to use all these -omic profiles to help reduce animal testing while improving early stage compound safety assessment and creating a public resource for the scientific community.”

“To improve safety testing, we need lots of data, and we’re grateful for our fellow consortium members who contribute resources, expertise, and experimental systems to the effort,” said Anne Carpenter, the public sector lead of OASIS and senior director of the Imaging Platform at the Broad, where she is also an institute scientist. “Only by working together can we achieve the shared goal of improving the way drugs are tested for safety.”

About a decade ago, Broad scientists including Carpenter developed a microscopy assay called Cell Painting, which uses fluorescent dyes to capture images of cells that reflect the cells’ response to treatments such as drug or genetic perturbations. It is already used throughout the pharmaceutical industry to discover the causes of and treatments for disease, but recently the team recognized the potential of Cell Painting, along with transcriptomic and proteomic methods, for toxicity testing. In 2023, they teamed up with HESI’s Emerging Systems Toxicology for the Assessment of Risk (eSTAR) Committee to secure funding from Massachusetts Life Sciences Center and launch the OASIS consortium.

During the consortium’s first three-year phase, members have identified roughly 1,500 compounds that have previously been tested for toxicity in animals or humans, including pharmaceuticals, industrial chemicals, and pesticides. They will then expose different liver cell models in the lab to those compounds and generate Cell Painting, proteomic, and transcriptomic data. They will use that data to train machine learning models to recognize profiles of cellular response to toxic compounds. The eventual goal is to run the same analyses on untested compounds and produce profiles that accurately predict whether those compounds would damage the liver in humans, while minimizing the need for animal tests.

One of the project’s first efforts was to locate, gather, and harmonize existing datasets from many different resources and to build partnerships with companies willing to share that data.

In addition, Broad scientists have used Cell Painting to measure the effect of their compound collection on two cell models, and they are starting to share those results with the scientific community. They’ve also shared work comparing machine learning-based models for predicting drug-induced liver injury. Other consortium members are using transcriptomic and proteomic assays to analyze the changes in proteins and gene activity after exposing cells to various compounds.

OASIS has also shared the list of compounds, so that groups or companies looking to get involved can easily procure them and launch experiments using their own cell models or assays and contribute data to the effort.

“The data we are generating from our first phase will build a strong foundation that’s based on as many different physiologically relevant models as possible and on a core set of compounds that we think are most informative,” said Singh.

Project leaders plan to complete data collection and then refine their machine learning models, in addition to welcoming more member groups that can contribute compounds, cell models, or assays. Members get access not only to valuable data, but also to this unique community, Singh said. “There’s a strategic value in being part of this scientific think tank, and there’s so much to be learned by shaping this effort together,” he added.