CellTICS: an explainable neural network for cell-type identification and interpretation based on single-cell RNA-seq data

Abstract

Identifying cell types is crucial for understanding the functional units of an organism. Machine learning has shown promising performance in identifying cell types, but many existing methods lack biological significance due to poor interpretability. However, it is of the utmost importance to understand what makes cells share the same function and form a specific cell type, motivating us to propose a biologically interpretable method. CellTICS prioritizes marker genes with cell-type-specific expression, using a hierarchy of biological pathways for neural network construction, and applying a multi-predictive-layer strategy to predict cell and sub-cell types. CellTICS usually outperforms existing methods in prediction accuracy. Moreover, CellTICS can reveal pathways that define a cell type or a cell type under specific physiological conditions, such as disease or aging. The nonlinear nature of neural networks enables us to identify many novel pathways. Interestingly, some of the pathways identified by CellTICS exhibit differential expression “variability” rather than differential expression across cell types, indicating that expression stochasticity within a pathway could be an important feature characteristic of a cell type. Overall, CellTICS provides a biologically interpretable method for identifying and characterizing cell types, shedding light on the underlying pathways that define cellular heterogeneity and its role in organismal function. CellTICS is available at https://github.com/qyyin0516/CellTICS.

GraSP Gene Targets to Hierarchically Infer Sub-Classes with CuttleNet

“…Hierarchical classification as a NN architectural strategy is in many ways another embodiment of the Coarse-to-Fine perceptual strategy. Specifically, many exciting projects have analogized this idea, by segmenting the task into steps they introspectively perceive their own mind is taking (Wang et al, 2012; Zhu & Bain, 2017; Sarlin et al, 2019), by directly trying to build a system in the style of empirically observed brain structures (Hafner et al, 2017), or even by mimicking behavioral strategies employed by by biologists (Yin & Chen, 2023).…”

“…The most similar project to our own from the perspective of hierarchical NNs is the CellTICS network by Yin & Chen (2023) which is also an explicitly hierarchical NN designed to distinguish cell classes and cell subclasses via supervised training with pre-labeled scRNAseq data. The authors demonstrate through extensive benchmarking that their cutting edge model is both the best architecture for this task published to date, and is by design interpretable.…”

“…We finally benchmarked our networks against CellTICS, the most cutting-edge published hierarchical NN model for simultaneous cell class and subclass classification (Yin & Chen, 2023). Appreciating the extensive benchmarking of CellTICS against other leading methods, and its superior performance, we deemed it the best reference for evaluating our models.…”

“…Here we bridge this gap by contributing: 1) A novel neural network (NN) training strategy for efficient dimensionality reduction. As well as, 2) CuttleNet, a hierarchical NN architecture inspired by coarse-to-fine biological perception, achieving state-of-the-art performance scores benchmarked against the best available supervised classifier for this problem (Yin & Chen, 2023); outperforming it on our specific application.…”

“…GraSP enables a 60-fold reduction in gene targets needed to classify all subclasses in the murine retina, outperforming expert-chosen and standard ML tools. CuttleNet achieves state-of-the-art performance for hierarchical cell class and subclass inference, surpassing the best available supervised classifier [7].…”