CellCodex has been established by an experienced team with a mission to provide better biological data to train advanced artificial intelligence models that could improve drug discovery.
Co-founded and led by CEO Dr Emmanouil Metzakopian, formerly the chief scientific officer at cell coding company bit.bio, the company has lab space at Babraham Research Campus and is engaged in talks with investors.
"There is a huge effort at the moment in the world to train artificial intelligence models to increase efficiency in drug discovery and reduce the cost of taking drugs to the clinic," Dr Metzakopian tells the Cambridge Independent. "However, the AI models out there at the moment have not really lived up to the promise. The reason for that is there hasn't been the right data out there for the models to be trained with.
"Iterations of artificial intelligence models have been moving in the speed of weeks and months whereas new data has been really lagging behind. It's causing a huge bottleneck in progress in this space and in applying AI and machine learning in drug discovery. There's a huge race in the world now to generate the right data and to train the right models so that we can fill this gap. CellCodex comes in exactly for that."
CellCodex creates customized human cell models with scalable production for predictive drug discovery and AI-driven biology.
Dr Metzakopian, who previously worked at the Wellcome Sanger Institute, focusing on gene editing and genetic screening, and at the UK Dementia Research Institute at the University of Cambridge, continues: "CellCodex produces not only the right kind of data but at the right scale. And the type of data that you need is perturbation data.
"By perturbation we mean we change the effect of a gene, or the expression of a gene, purposefully - either activate it or repress it - and we see how a human cell responds. We take that into account in basal positions but also in disease positions. How does the human cell respond to the changes in the gene?
"We do that for all the genes in the genome in millions and millions of experiments and generate deep multi-omic data covering, in simple terms, many features of a cell - gene expressions of all genes in the genome, or cellular behaviour and multi-cellular interactions as well."
Data is captured at the molecular, cellular and multicellular levels using perturbation components that including genetic data, high-throughput CRISPR edits, iPSC-derived cell types, complex in vitro models, environmental exposures and single-cell readouts/imaging.
The data is then supplied to partners looking to train their AI models.
"We do it in the right quality, with the right metadata - all the information that encompasses the experiment, from the operator to the equipment used to the quality along the process," says Dr Metzakopian. "We bridge biology - the best human models - with a functional genomics approach to peturb genes and with our knowhow we generate the data in the right format to train machine learning models."
Dr Metzakopian's co-founders include computational biologist Dr Grant Belgard, the chief technology officer responsible for building the platform that prepares the large-scale gene perturbation data for training virtual cell models. He founded and leads The Bioinformatics CRO, which is part of CellCodex's platform, and was vice-president of bioinformatics at bit.bio.
Fellow co-founder Dr Tom Weaver is chairman at CellCodex. An experienced scientist, serial entrepreneur and angel investor specialising in genetics, genomics and drug discovery, he founded and led companies including Geneservice, Congenica and PetMedix - the latter was acquired in 2023 by Zoetis.
The fourth co-founder is Ravi Moorthy, chief communications officer, who was previously vice president of communications at Scorpion Therapeutics, a precision oncology biotech start-up that was acquired by Eli Lilly.
Completing the team of six are head of functional genomics Dr Andreas Andreou, who is also co-founder and chief scientific officer at Prozymi Biolabs, and Eve Coomber, who is the CellCodex's head of stem cell biology, and spent a decade at the Wellcome Sanger Institute, specialising in genetic engineering and developing new models, tools, and screening techniques.
Clients for CellCodex include tech companies building AI models, along with pharmaceutical companies. But what makes CellCodex's offer the one they would choose?
"Two things: one is the biology," replies Dr Metzakopian. "We have a background in developing the right human cells that have high fidelity to the cell types in our bodies. And we can model disease accurately.
"Secondly, we can perform the right quality checks along every step of the process to make sure there's no biological or methodological drift. That maintains the quality of the data and its fidelity with regards to human health.
"Depending on the requirements and the space that the partner is working in, we would use either single cell types, complex cultures or organoids and we have experience across the spectrum."
CellCodex is working on a funding round.
"We are engaging with the Cambridge ecosystem and with venture capitalists and speaking to a few angels. It is progressing really well," says Dr Metzakopian.
Establishing his own company "feels fantastic", he says.
"Finding a gap that is so important to fill and finding a way to do it feels amazing," he adds. "Now we need to execute and deliver on that promise. And I can't wait to be part of this initiative. If our datasets are enabling models that lead to better drugs, I'll be the happiest person in the world."