Diseases evolve and manifest due to accumulation of genetic and epigenetic dysregulations over our lifetime. Through my Ph.D. work, I proposed a new model, the disease-state trajectories (DST), which explains how diseases might evolve (Figure.1).

Figure 1: The disease-state trajectories (DST) model and dysregulated-cellular states. The greenish colors represent the cellular state in healthy individuals. A dysregulated cellular state (yellowish color) starts to perturb the healthy trajectory that would lead to the emergence of the disease-state trajectories (DST) (red colors).  

This model is based on extensive analyses of many tumour entities at single-cell resolution. The DST model proposes that diseases evolution is a continuum process, and most likely, our patients have to go through sequential steps of certain events, passing by “Dysregulated-State” till the manifestation of full-blown “Disease-State” trajectories.

The key essence of our biological research is to define the underlying causal paths which describe the governing mechanisms of the disease phenotypes. Unfortunately, we are not able to have access to the full continuum process, the intermediate states, since such processes evolve over many years and sampling these processes is quite challenging. This led to the question, can we design AI agents which are able to reason on top of the data and reconstruct the intermediate-cellular states?

Over my postdoctoral research,  I developed Cell-DRL to tackle this question. Cell-DRL is a deep reinforcement learning agent able to generate actions in the gene-expression space and learn stochastic policies predicting trajectories and causal paths which connect two anchoring cellular states. We expect that the Cell-DRL approach will be crucial to capture the underlying causes of diseases at single-cell resolution.