12 of the top 20 best-selling drugs are monoclonal antibodies, generating over 60% of total drug revenue. A keyword search on PubMed for ‘antibody’ ‘machine learning’ or ‘deep learning’ generates many results. There has been a steep increase in the number of publications on these topics over the years. Therefore, there is huge interest and investment in this modality but the time to market is extremely long and expensive.
According to Jannick Bendtsen, CEO & Co-Founder of PipeBio, there has been a shift from wet lab to dry lab interest in machine learning. PipeBio is a bioinformatics cloud platform that includes high-flow mappings such as LIBRA-seq or BEAM from 10X Genomics used to conduct large-scale sequence analysis of antibodies. A dry lab generates high volumes of assay data and there is demand for methods that automatically capture this data in a structured way. This task is challenging due to data volume and the various modalities generating heterogeneous data.
Bendtsen highlighted that there is no linear path in antibody discovery; there is a lot of back on forth as new models and data are generated. Benchling has joined forces with PipeBio to create databases in a structured manner. Using the registered data, scientists can begin designing their bi-specifics or more advanced complexes and scaffolds.
PipeBio’s main focus is on sequence analysis of antibodies but since partnering with Benchling company they have expanded their range of capabilities from end-to-end data capture, analysis, and registration from raw sequence data to in vivo testing. The company is also working to improve its developability prediction tools. So, they have created developability charts or metrics where one can observe individual antibodies in high throughput, analyse sequences, and determine where they will fit into a developability scale.
Visualisation tools help discern certain trends in the data, (data points can be overlaid with existing graphics). For wet lab scientists, writing SQL queries may not be in their wheelhouse, therefore the team at PipeBio developed a large language model to assist with this. All the user has to do is propose the SQL syntax and the model will generate a dashboard which facilitates data visualisation.
Bendtsen gave an insightful overview of the current hurdles in therapeutic antibody development along with the tools available that could provide potential solutions. By teaming up with Benchling, PipeBio has strengthened its capabilities to capture, automate, analyse, and present data in a comprehensive manner for customers.
