Theoretical foundations for the Human Cell Atlas

TL;DR

This paper develops a mathematical framework based on sheaves of modules to determine when biological experiments can be reliably mapped to each other, aiding the analysis and integration of complex datasets like those in the Human Cell Atlas.

Contribution

It introduces a formal sheaf-theoretic approach to assess experimental equivalence and phenotypic identifiability in biological data analysis.

Findings

Provides conditions for experimental mapping based on sheaf structures

Formalizes biological assumptions within a mathematical framework

Enables design of reliable complex biological experiments

Abstract

In Schiebinger et al. (2017), the authors use optimal transport of measures on empirical distributions arising from biological experiments to relate the single cell RNA sequencing profiles for induced pluripotent stem cells differentiating. But such algorithms could be arbitrarily applied to any datasets from any collection of experiments. We consider here a natural question that arises: in a manner consistent with conventionally accepted assumptions about biology, in which cases can the results of two experiments be mapped to each other in this manner? The answer to this question is of fundamental practical importance in developing algorithms that use this method for analysing and integrating complex datasets collected as part of the Human Cell Atlas. Here, we develop a formulation of biology in terms of sheaves of $C^*(X)$-modules for a smooth manifold $X$ equipped with certain structures, that enables this question to be formally answered, leading to formal statements about experimental inference and phenotypic identifiability. These structures capture a perspective on biology that is consistent with a standard, widely accepted biological perspective and is mathematically intuitive. Our methods provide a framework in which to design complex experiments and the algorithms to analyse them in a way that their conclusions can be believed.