Treatment with immune checkpoint inhibitors (ICIs) is becoming the systemic “standard of care” for an increasing number of cancer patients. The purpose of this treatment is boosting the host immune system to destroy the cancer cells but, unfortunately, it is very costly and only a fraction of patients (20-40%) show response. Accurate and rapid prediction of response could avoid undergoing this costly treatment unnecessarily while enabling oncologists to choose a more suitable option as early as possible. Furthermore, there is compelling evidence that monitoring changes in immune cell populations during the course of a treatment could be used as a proxy for response to ICI. Therefore, an accurate estimation of the immune cell proportions over time is crucial. By integrating transcriptomics and immune cell profiling of blood (plasma, platelets and PBMCs), stool metagenomics and formalin-fixed paraffin-embedded tumor samples from a unique sample cohort of 97 cancer patients undergoing immunotherapy with ICIs (at baseline and at several time points during the course of the treatment), I aim to establish a time-course multi-omics framework to predict response to ICI and a complementary computational deconvolution pipeline to accurately estimate the proportions of different immune cell populations across complementary blood fractions.