Applying Artificial Intelligence to Biomanufacturing

28 April 2021

Stain-free image analysis

Artificial Intelligence [AI] is everywhere. Each time we unlock a phone or speak with a virtual assistant, we are interacting with AI. It shields us from spam emails, protects our bank accounts and helps secure our computers. AI even suggests new songs we might like and what movies to watch next. From self-driving cars to flying autonomous drones on Mars, AI is already a major part of our world.

In manufacturing, AI coupled with Internet of Things [IoT] sensors and Digital Twins (process models), is driving a fourth Industrial Revolution offering higher yields of better quality products from more efficient processes. This approach already dominates car assembly lines and computer chip manufacturing, but has not yet disrupted biomanufacturing. ValitaCell are developing AI tools to drive next-generation biopharmaceutical manufacturing.

The Challenge

Biopharmaceuticals are the fastest growing and largest segment of the pharmaceutical industry. Biologics are protein drugs manufactured inside living cells. These account for some of our most clinically and commercially important medicines. In the rapidly emerging Cell Therapy area, the cells themselves are the medicine administered to the patient.

Manufacturing within or of cells is extremely complicated and can easily go awry. Yet throughout biomanufacturing design and operation, the sector does not have fast, rich analytical tools to monitor how cells are coping with the stresses and strains of production. Without these cell analytics we cannot react to in-process issues to keep manufacturing on track. A fast, rich cell analytic is vital for biomanufacturing to join the fourth Industrial Revolution.

To analyze cells in the laboratory we often use staining to tease information out of images about cell health and quality. Yet staining is too slow and difficult to implement on the factory floor. Tantalisingly, experts often can spot key information in unstained images. In such cases stains are used to simplify image processing, but if we could train computers to emulate the expert’s eye then we could do away with stains.

The opportunity

ValitaCell are training AI models to do exactly this. Our Cloud-based AI algorithms – called CellAi – are built to interpret unstained cell images to analyse cells quickly, in-depth and without destruction. We do this by feeding stain-free input images and target assay results, or pairs of input unstained and output stained images, into AI models. The AI training process then learns how to map from inputs to outputs. Then, when fed with an unstained input image only, the model can predict the output without staining.

Eliminating staining means we can analyse cells faster, at lower cost and without wasting precious sample. Additionally, in the laboratory we can apply at most 4-5 stains per sample, but with CellAi’s virtual staining there is no limit on multiplexing – we can run any number of models on the unstained image. Requiring only stain-free imaging and software, the CellAi approach also readily integrates into automation workflows. The CellAi approach even helps cut laboratory plastic waste.

Predicting CHO cell viability

One CellAi tool quantifies Chinese hamster ovary cell viability without staining. To build this we used Deep Learning to map from brightfield cell images onto propidium iodide stains. Now, with only an unstained brightfield image, we are able to predict whether a cell is alive or dead (and so calculate a cell population’s percentage viability) without any staining.

Image A: Truth
Image B: Prediction

CHO cell viability prediction. Left: a false colour rendering of a 10x brightfield image overlaid with actual (binarized) propidium iodide staining. Right: CellAi prediction of PI staining.

Virtual Staining

In our DeepStain project, we have built models that map from unstained mesenchymal stem cell images onto nuclear stains. Nuclei are rather tricky to spot in brightfield images, yet the models manage to replicate staining. As unstained images can be generated very quickly, we have integrated this “virtual staining” into our media development platform to increase screening throughput.

Virtual nuclear staining of mesenchymal stem cells. Top-left: 20x brightfield MSCs. Top-right: Hoechst nuclei staining. Bottom-left: Virtual nuclear staining. Bottom-right: Virtual stain overlaid on brightfield image.

Moving ahead

CellAi will simplify, accelerate and reduce the cost of biomanufacturing. It will do this by much more than making current processes simply run faster because, by plugging the gap around cell analytics, it will enable manufacturers to develop smarter, leaner factories. We are already embedding CellAi into in-house and partner Cell Line Development, Media Development and automated manufacturing platforms. automated manufacturing platforms.

We continue to refine and optimise our CellAi algorithms and technology and welcome conversations with potential partners on how we can solve biomanufacturing problems together.

Dr. Paul Dobson

Head of Data

Dr. Dobson is a biochemist by degree with a PhD in Machine Learning. He has worked for 20 years at the intersection of Biology, Computer Science and Engineering. He has published 35 peer-reviewed research articles, reviews and book chapters across bioinformatics, scientific text mining, drug discovery, systems biology and bioprocess engineering. At ValitaCell, Paul leads the Data Team, applying Machine and Deep Learning across the company's portfolio of analytical tools that support better biologics and cell therapy manufacturing.

Dr. Ben Thompson

Chief Technology Officer

Dr. Thompson is a cell line development scientist with a unique combination of education and expertise in bioprocess engineering, mathematics and statistics. Previously, he held the position of Postdoctoral Research Fellow at the University of Sheffield. Dr. Thompson has a Ph.D. in Bioprocess Engineering and Graduate Certificate in Statistics [Distinction]. He is a key inventor on a number of ValitaCell patents (Fluorescence Polarisation quantitation, Nanobody Quantitation) and he is a domain AI expert with patents pending [Cell Biology].