1438 Integrative analysis of single cell multiomics data using deep learning to identify immune related biomarkers in a patient derived 3D ex vivo tumoroid platform

Background

Our previous studies demonstrated that multiple types of omics data obtained from Nilogen’s comprehensive 3D-EXplore ex vivo drug testing platform using tumoroids with intact tumor microenvironment prepared from unpropagated fresh patient tumor samples can reveal cellular mechanisms that are active in individual tumors. This approach allows classification of tumors into subtypes for response to immunotherapeutic drug treatments. Here we aimed to utilize Scailyte’s cluster-free, unbiased, and highly sensitive AI platform ScaiVision to better understand the cellular and molecular mechanisms of cGAS-STING activation alone or in combination with an anti-PD-1 treatment to discover clinically relevant biomarker signatures.

Methods

For the 3D-EXplore ex vivo platform, tumoroids measuring 150 µm in size were prepared from procured fresh tumor tissue of renal cell carcinoma (RCC) and colorectal carcinoma (CRC) patients. The study was approved by the Institutional Ethics Board, approval number Pro00014313. Tumoroids were generated using a proprietary mechanical process without any enzymatic digestion or propagation and were treated ex vivo with STING agonists ADU-S100 and 2’3’cGAMP alone or in combination with nivolumab. Treatment-mediated changes in the tumor immune microenvironment were analyzed using Cellular Indexing of Transcriptomes and Epitopes by sequencing (CITE-seq), multicolor flow cytometric analysis, and multiplex cytokine release assays.

Results

Here we present the results from the integrative analyses of multiomic data generated from the treatment of fresh patient tumor samples in the ex vivo assays. To identify cellular and molecular mechanisms associated with ex vivo responses to the cGAS-STING pathway activation and PD-1/PD-L1 checkpoint blockade therapy we used a supervised machine learning algorithm (ScaiVision) that trains a convolutional neural network with a single layer to predict sample-level labels using single-cell data as inputs. This approach allowed an in-depth characterization of the tumor resident immune cell subsets in the patient tumoroids before and after ex vivo drug treatment to reconstruct cell-type-specific signaling responses to identify cell populations associated with response to immunotherapeutics ex vivo.

Conclusions

Our results revealed that the ScaiVision platform allows the integration of high-parameter single-cell data together with multiple types of omics data derived from Nilogen’s 3D-EXplore platform to better understand molecular and cellular mechanisms of drug mode of action that may allow the discovery of new clinically-relevant immune-related biomarkers to predict clinical outcome of immunotherapy.

Ethics Approval

All tissues were collected under proper patient consent and the study approved by the Institutional Ethics Board, approval number Pro00014313