CanadaHealthcare and life sciences2020 – ongoingAbCellera
Machine-learning antibody discovery for the COVID-19 programme with Eli Lilly
AbCellera and Lilly screened 5.5 million cells in three days and narrowed 530 COVID-19 antibodies to 24 in 23 days
Problem
Traditional hybridoma campaigns take two to three months to find a handful of antibodies. The COVID-19 programme with Eli Lilly needed suitable candidates from a survivor's blood sample much faster.
Approach
AbCellera ran its own platform on a sample from a COVID-19 survivor. A microfluidic device with 200,000 miniaturised chambers and machine-vision imaging picked out antibody-secreting cells. Machine learning on genetic sequences, trained on antibodies with more than 100 properties each, looked for patterns linked to potency and manufacturability, and its Celium software grouped antibodies into networks for scientists to choose from. The company says AI output must be checked against lab work.
Results
Takeaway
The speed came from pairing imaging hardware with models trained on years of antibody data. The company treats the AI as a tool whose picks are validated in the lab.
Categories
- Industry
- Healthcare and life sciences
- Business function
- IT and R&D
- Innovation type
- Product and service innovation
- AI technology
- Computer visionPredictive analytics and ML
- Adoption
- Built in-house
- Evidence
- Press reporting
Sources
Source summaries
BioSpace describes how AbCellera used machine vision and machine learning to find COVID-19 antibodies for its programme with Lilly. It reports 5.5 million cells tested in three days, 530 unique antibodies, and 24 candidates chosen in 23 days. It also gives a target of four months from screening to first dosing in humans, a plan rather than a result.
Full details are in the original sources.
