Table 1:

Computational techniques for dimensionality reduction and mechanistic understanding of single-cell heterogeneous signaling data

Method	Output	Benefits	Shortcomings	Software	Applications
Clustering	Labels for each cell corresponding to cluster belonging.	Aggregates based on real cell behavior, relatively fast, interpretable.	Inappropriate or unhelpful for continuous responses	Time course inspector R package and web application [35]	[22], [34], [47]
FPCA	Low-dimensional projection of each cell along axes of variation in the data.	Enables easy visualization of heterogeneity to identify patterns in data.	Primary axes of variation may not be physiologically meaningful.	FDA package [36,37]	[38]
Information Theory	Numerical quantification of mutual information between input (dose or ligand) and output (reporter activity)	Summarizes heterogeneous outcomes in a single number with a physical meaning.	Mutual information may be of limited interest for particular study.	SLEMI R package [59], EstCC Scala package [60]	[11,39–42,44]
Stochastic Dynamic modeling	A distribution of species dynamics corresponding to each parameter set used by modeler.	Can explicitly model the effects of intrinsic stochasticity and low molecular number.	Requires knowledge of rate constants and species interactions, along with simplifying assumptions to make system tractable.	Simbiology MATLAB package, Hy3S [61], PySB Python package [43]	[43,44,46]
ODE modeling	A set of deterministic species dynamics corresponding to each parameter set used by modeler.	Can vary parameters to test hypotheses about sources of heterogeneity, suggest mechanistic drivers.	Requires knowledge of rate constants and species interactions, along with simplifying assumptions to make system tractable.	Simbiology MATLAB package, PySB Python package [43]	[11,16,43,47,48]