TxCyto: A machine learning framework for estimating cytokine activity from whole transcriptome

Cytokines play a vital role in cell-to-cell communication, and comprehensively understanding their activity is crucial for unraveling health and disease complexities. Current methods to estimate cytokine activity depend on experimental measurements, which are only available for a limited set of cytokines (43) and context-specific, making comprehensive profiling across different tissues, disease states, and biological contexts unfeasible.

To tackle this challenge, researchers have created TxCyto, an innovative machine learning framework that estimates cytokine activity directly from the whole transcriptome profile of a sample.

TxCyto was trained on pan-cancer TCGA tumor transcriptomes and has been rigorously evaluated using multiple independent datasets, including cytokine perturbation experiments. The framework has demonstrated its effectiveness across various cancer immunotherapy cohorts, successfully identifying cytokines whose predicted activity correlates with therapeutic response. Additionally, when applied to spatial transcriptomic data for Liver cancer, TxCyto unveiled spatial niches linked to immunotherapy response.

In summary, TxCyto is a powerful machine learning tool designed to predict the activity of 645 cytokines and the broader tumor secretome from readily available whole transcriptomes. This versatile framework can be adapted to other classes of regulatory molecules, and the codebase and tools are accessible at https://github.com/Rahulncbs/TxCyto.