Abstract 1932: Pollock: Fishing for cell states

The use of single-cell methods is expanding at an ever-increasing rate. While multiple algorithms address the task of cell classification, they are limited in terms of cross platform compatibility, reliance on the availability of a reference dataset, and classification interpretability. Here, we introduce Pollock, a suite of algorithms for cell type identification that is compatible with popular single cell methods and analysis platforms, provides a series of pretrained human cancer reference models, and reports interpretability scores that identify the genes that drive cell type classifications. Our model combines two important approaches, one each from machine learning and deep learning: a variational autoencoder (VAE) and random forest classifier, to make cell type predictions. Pollock is highly versatile, being available as a command line tool, Python library (with scanpy integration), or R library (with Seurat integration), and can be installed as a conda package, or in containerized form via Docker. To allow for easier pan-disease and pan-tissue analyses, Pollock also ships with a library of pretrained cancer type specific and agnostic modules that were trained on expertly-curated single cell data that are ready to “plug and play” with no additional annotation or training required. Conversely, Pollock also allows for the training of custom classification modules, if an annotated reference single cell dataset is available. These pretrained models were fitted on manually curated and annotated single cell data from eight different cancer types spanning three single cell technologies (scRNA-seq, snRNA-seq, and snATAC-seq). Pollock also provides feature importance scores that allow for cell type classifications to be traced back to the genes influencing a particular cell type classification, further promoting biological interpretability. These scores could allow for new, technology-specific biomarker discovery. We also demonstrate the utility of Pollock by applying it in a pan-cancer single cell immune analysis.

Citation Format: Erik Storrs, Daniel Cui Zhou, Michael C. Wendl, Matthew A. Wyczalkowski, Alla Karpova, Liang-Bo Wang, Yize Li, Austin Southard-Smith, Reyka G. Jayasinghe, Lijun Yao, Ruiyang Liu, Yige Wu, Nadezhda V. Terekhanova, Houxiang Zhu, John M. Herndon, Feng Chen, William E. Gillanders, Ryan C. Fields, Li Ding. Pollock: Fishing for cell states [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1932.

Pollock: fishing for cell states

Motivation

The use of single-cell methods is expanding at an ever-increasing rate. While there are established algorithms that address cell classification, they are limited in terms of cross platform compatibility, reliance on the availability of a reference dataset and classification interpretability. Here, we introduce Pollock, a suite of algorithms for cell type identification that is compatible with popular single-cell methods and analysis platforms, provides a set of pretrained human cancer reference models, and reports interpretability scores that identify the genes that drive cell type classifications.

Results

Pollock performs comparably to existing classification methods, while offering easily deployable pretrained classification models across a wide variety of tissue and data types. Additionally, it demonstrates utility in immune pan-cancer analysis.

Availability and implementation

Source code and documentation are available at https://github.com/ding-lab/pollock. Pretrained models and datasets are available for download at https://zenodo.org/record/5895221.

Supplementary information

Supplementary data are available at Bioinformatics Advances online.

Abstract PO-051: Tumor stem cells arising from a non-stem origin maintain a differentiated phenotype and modulate T cell activity

Tumor stem cells (TSCs) contribute to cancer mortality via therapeutic resistance, tumor recurrence, and metastatic mechanisms. However, the origins of the stem capacity to TSCs remains in question, but all TSCs descend from the original tumor cell-of-origin where the first oncogenic event occurred. Tumors arising from different cells-of-origin are histologically identical, but it is unknown whether TSCs that arose from different origins are molecularly and functionally distinct. Using mouse models driving identical Apc mutations from Lrig1-expressing and Mist1-expressing cells, we characterized TSCs of tumors driven from stem and non-stem cells-of-origin using single cell RNA sequencing (scRNA-seq), organoids, and multiplexed imaging. We revealed reduced stem capacity but increased class II antigen presentation ability for non-stem cell (Mist1) driven TSCs compared with stem cell (Lrig1) driven TSCs, which resulted in a favorable immune microenvironment skewed towards active cytotoxic response in Mist1-driven tumors. These results suggest that the cell-of-origin of tumorigenesis provides a specific context by which TSCs are generated, dictating their interactions with the tumor microenvironment.

Citation Format: Cherie’ R. Scurrah, Bob Chen, Nick Markham, Alan Simmons, Austin Southard-Smith, Mary Macedonia, Eunyoung Choi, Qi Liu, Kay Washington, Bob Coffey, Jeremy Goettel, Ken Lau. Tumor stem cells arising from a non-stem origin maintain a differentiated phenotype and modulate T cell activity [abstract]. In: Proceedings of the AACR Virtual Special Conference on Tumor Heterogeneity: From Single Cells to Clinical Impact; 2020 Sep 17-18. Philadelphia (PA): AACR; Cancer Res 2020;80(21 Suppl):Abstract nr PO-051.