Researchers at Amsterdam UMC Cancer Center Amsterdam have developed a new machine learning tool, named Statescope, that can “listen in” to a tumors’ cell composition using standard sequencing data in an unprecedented manner. With Statescope the researchers identified a specific combination of active immune cells that can predict a patient's benefit from immunotherapy. These findings by Jurriaan Janssen, Mischa Steketee, Teodora Radonic, Bauke Ylstra, Yongsoo Kim and colleagues have been published in Nature Communications.

Decoding cellular composition and activation states

Standard bulk RNA- and DNA-sequencing data from tumors are widely available in academic research institutes and from public datasets. These bulk sequencing measurements reflect a mix of signals from all cells of the tumor, diluting the activation signals of individual cells, including malignant and immune cells. Statescope enables the disentanglement of these admixed signals into individual cell types and their corresponding states 

Immune cell states predict immunotherapy benefit

The team first built a detailed cell state atlas from nearly 1,000 lung cancer tumors (TCGA NSCLC) and over a hundred pancreatic cancer tumors (TCGA PAAD). This atlas describes the many states of malignant cells and the cells in their direct (immune) environment. 

They subsequently applied this atlas with Statescope to deconvolute the cellular composition of several independent advanced lung cancer cohorts, including the phase II/III POPLAR/OAK trials, in which patients received either atezolizumab (a PD-L1 inhibitor) or docetaxel (chemotherapy). The research team successfully decoded not only the cellular composition of the immune cells in the tumor, but also their specific activation states, providing an unprecedentedly detailed view of the tumor immune environment that enabled the identification of immune cell combinations predictive of immunotherapy response. If a tumor is characterized with this active immune cell combination, the patient is much more likely to respond well to immunotherapy, but not to chemotherapy. 


"Statescope shows that we can “listen in” on the immune system of tumors using data that are nowadays generated for many studies and clinical trials , and use that information to move closer to the right treatment for the right patient" - researcher Yongsoo Kim

Read the full publication in Nature Communications here.