Historically, biological research and specifically drug discovery have relied on the analysis of population level information - from wide population-level studies in clinical trials, to examining organism-level and organ-level endpoints in preclinical animal studies, to identifying compound hits on whole-well level data in the in vitro stages of drug discovery. While approaches like these have been important for pinpointing major issues like toxicity endpoints and determining overall efficacy, until recently we have not been able to examine changes in individual cells at high-throughput screening scale. Advances in computing technology including the use of GPUs and the advent of artificial intelligence (AI) and  machine learning (ML) approaches have made feature extraction and data parsing of these complex data sets tenable. 

The Essence of Single Cell Analysis

Single cell analysis represents a paradigm shift in biological research and drug discovery, moving beyond the limitations of bulk analyses that average out the unique characteristics and responses of individual cells within a population. Single cell analysis acknowledges the profound heterogeneity among cells, even those of the same type, within tissues and organs, and even in in vitro cell cultures. By isolating and examining cells on an individual basis, researchers can uncover the nuanced genetic, proteomic, and metabolic differences that drive diverse biological processes and disease states.

The Problem with Averaging

In the case of high content imaging and phenotypic profiling, images of cells are captured following exposure to experimental conditions, the cells are segmented and features (measurements) are extracted. In most workflows, these measurements are then averaged on a field or well level and then analyzed over replicates to determine if a given experimental condition induces a response. These approaches ignore the inherent variability in responses from cell to cell, even in the most uniform samples. 

For cell types that are more biologically diverse like patient derived samples including peripheral blood mononuclear cells (PBMCs), this bulk analysis of responses leads to “noisy” data, and overlooks interesting, subtle phenotypic changes in individual cells. PBMC cultures include many different immune subtypes including lymphocytes (T cells, B cells, and NK cells), monocytes, and dendritic cells. Being able to tie a response to a change in a specific PBMC subpopulation can provide important information about the mechanism of action of a given compound or condition. 

Enhancing Data Analysis

One of the primary challenges in single cell analysis is the sheer volume and complexity of the data generated. Each cell can provide a wealth of information, from gene expression profiles to dynamic responses to environmental stimuli. AI algorithms excel at sifting through this multiparametric data, identifying relevant patterns, and distinguishing between different cell types, states, and functional characteristics, but this can come at significant computing cost. 

One modern way to analyze single cell data is to develop supervised machine learning classifiers, where scientists can train a classifier based on a known phenotype. For example, in a study examining inflammasome biology, inflammasome activation is characterized phenotypically by the development of an ASC speck. Scientists can use this known prior and train a machine learning classifier for that specific phenotype, and then examine how this population is changing across conditions. Using Spring Engine, this can be done using PhenoSorter, as shown in the video below.