Early detection of side effects through AI-supported video analysis

The customer

Boehringer Ingelheim is a biopharmaceutical company active in both human and animal health. As one of the industry’s top investors in research and development, the company focuses on developing innovative therapies that can improve and extend lives in areas of high unmet medical need. Independent since its foundation in 1885, Boehringer takes a long-term perspective, embedding sustainability along the entire value chain. More than 53,500 employees serve over 130 markets.

The challenge

This collaboration focused on enhancing pre-clinical safety assessments, which are critical early steps in drug development. The challenge involved pre-clinical safety studies conducted on rodents, which are mandatory before advancing to human trials.

These studies aim at detecting unforeseen side effects early on and produce large video datasets of rodent behavior, which previously could be analyzed for a few behavioral parameters, increasing the risk of missing relevant behaviors. To improve the efficiency and precision of the studies, Boehringer Ingelheim was looking for a research partner through a public call on its open innovation portal opnMe.com to develop an AI-driven application for analyzing these behavioral patterns.

"Merantix Momentum quickly and precisely understood our needs and translated them into a highly innovative and user-friendly machine learning solution. The tool enables us to perform significantly more detailed video analyses, faster, and contributes to the safety in drug discovery"

Director General Pharmacology, Boehringer Ingelheim GmbH & Co. KG

Solution

Among 27 contestants that participated in the call, Merantix Momentum stood out to the committee of data science and pharmacology experts with a user-centric and technologically robust approach, prioritizing human oversight and control. As part of the joint research effort, Merantix Momentum developed a software solution to label and analyze rodent behavior focusing on safety relevant tests for potential novel drug candidates.

Leveraging our expertise in end-to-end machine learning solutions, we employ advanced computer vision techniques to identify new movement patterns and classify known behaviors. Created in collaboration with safety pharmacology scientists, the tool offers novel ways to examine behavioral patterns and conduct detailed analyses of potential substance-related effects. Additionally, the solution uses an Active Learning approach to continually enhance the model by learning efficiently from user interactions.

Results & effects

Accuracy: More than 20 new behavior types are classified with 87% accuracy and unusual or ambiguous behaviors are highlighted, which improves the detection of potential adverse effects.

Autonomy: Due to the Active Learning approach, scientists were able to increase the amount of detected behavior types tenfold and improve the machine learning model without support from machine learning engineers.

Acceleration: New safety studies, can be analyzed in a day whereas previously detection of these novel behaviors was not possible at all.

The collaboration with Boehringer Ingelheim enhances analysis and interpretation of video-based safety assessments thereby contributing to drug safety during the drug discovery process. The use of advanced AI tools has significantly improved the efficiency of analyses providing completely novel parameters and tools to assess drug safety. These advancements contribute to fast and safe delivery of new therapies to the market.

Read more about our work in our paperthat we published at ICML 2024 and Measuring Behavior 2024.

‍