Using AI to expand the universe of cell types

In this proposal, we will utilise advances in artificial intelligence (AI) to open a new area of cell biology by creating novel cell types that do not naturally occur but can be programmed to display chosen functions. This will be achieved by creating generative AI tools that can learn the characteristics that define the gene expression profile of naturally occurring cells from the abundance of single-cell data that is now available. Once learnt, these characteristics will be used to create a latent space that represents all plausible cell types, including those which are not naturally occurring. By developing the models to include the functional characteristics of the naturally occurring cells by incorporating conditional information from the existing functional ontologies, it will be possible to navigate this latent space according to which functions that region of the space is likely to encode. As a result, it will be possible to generate artificial gene expression profiles for novel cell types that we will refer to as In-silico derived cells (or IDCs). The gene expression profiles from IDCs and gene expression profiles of a naturally occurring cell type (a human embryonic stem cell) will be used as input to existing algorithms that have already been demonstrated to predict what is required to inter-convert between cell states. These algorithms will provide predictions for both the sets of regulators to over-express in the naturally occurring cell types to push them towards the IDC state and the culture conditions required to maintain IDCs once converted. To demonstrate the validity of this approach, we will experimentally validate the predictions using a CRISPR-activation screening platform that we have previously developed. As a result, we will collect gene expression profiles of cells transformed from a naturally occurring to an IDC cell state and compare them with the predicted gene expression profiles. This will prove that the creation of IDC states is possible, forcing us to reconsider our understanding of cell type, the flexibility of the genome and the extent to which cell-state engineering can be used to create novel cell therapy treatments.