We Help Make Science More Successful

The first reagent we started with is antibodies.

About half a million scientists spend over $3 billion a year buying research antibodies for experiments. They use these to detect and quantify proteins.

Selecting antibodies slows the velocity of research and drug development. There are more than 6.6 million commercial antibodies, and vendors can't predict how each one will work in specific experiments. Data on antibody use is buried in biomedical papers, vendor catalogs, and independent validation databases. In part because this data is hard to find, up to 50% of selected antibodies don't work in experiments†.

This wastes resources and delays research projects. Our work with pharmaceutical companies suggests each spends $1-3 million a year on commercial antibodies that don’t work. In addition, researchers:

• Purchase unnecessary custom antibodies for $50,000 that take up to 3-6 months to develop
• Spend days to select and weeks to test and validate antibodies
• Redundantly validate antibodies that colleagues in the same organization have already validated

Data to solve the problem is buried in scientific publications. Technological advances now allow it to be decoded. This includes advances in machine learning to better interpret text and images, increases in processing power to perform this analysis at scale, and improvements in graph databases to map and extract insights from results.

This has enabled huge benefits from a technology that we call AI-Assisted Antibody Selection. It can:

• Reduce the hard cost of consumables up to $3 million per year
• Accelerate projects by selecting antibodies in 30 seconds vs 12 weeks
• Empower organizational purpose by reducing manual searches and restoring research time
• Provide a turnkey application of AI to increase organizational efficiency

BenchSci is a pioneer in using AI to help scientists plan and select reagents for more successful experiments. We:

• Collate the world’s largest collection of antibody-specific data
• Identify biological entities in their text and images with proprietary machine learning models
• Map their relationships in a knowledge graph using bioinformatics databases and ontologies
• Provide an intuitive interface to select antibodies by protein target and experimental variables