# Evaluating transportability of in vitro cellular models to in vivo human phenotypes using gene perturbation data

## TL;DR

This paper introduces a framework to assess how well lab-based cell experiments predict real human biology using gene disruption data.

## Contribution

The novel GPAT framework evaluates transportability of in-vitro models to in-vivo human phenotypes using gene perturbation data.

## Key findings

- GPAT found evidence that lysosomal cholesterol accumulation in vitro relates to lower LDL-cholesterol in humans.
- Transportability was observed between cancer cell proliferation and human plasma cellular phenotypes.
- Some in-vitro models showed limited transportability to in-vivo human biology.

## Abstract

Gene perturbation screens (e.g. CRISPR-Cas9) assess the impact of gene disruption on in-vitro cellular phenotypes (e.g., proliferation, anti-viral response). In-vitro experiments can be useful models for in-vivo (organismal) phenotypes (e.g., immune cell anti-viral response and infectious diseases). However, assessing whether an in-vitro cellular model effectively captures in-vivo biology is challenging. An in-vitro model is ‘transportable’ to an in-vivo phenotype if perturbations impacting the in-vitro phenotype also impact the in-vivo phenotype with mechanism-consistent directionality and effect sizes. We propose a framework; Gene Perturbation Analysis for Transportability (GPAT), to assess model transportability using gene perturbation effect estimates from perturbation screens (in-vitro) and loss-of-function burden tests (in-vivo). In hypothesis-driven analyses, GPAT provides evidence for model transportability of higher lysosomal cholesterol accumulation in-vitro to lower human plasma LDL-cholesterol (P = 0.0006), consistent with the known role of lysosomes in lipid biosynthesis. In contrast, there was limited evidence for other putative in-vitro models. In hypothesis-free analyses, we find evidence for transportability of cancer cell line proliferation to in-vivo human plasma cellular phenotypes (e.g. erythroleukemia proliferation and plasma lymphocyte percentage). Here we show that perturbation data can be used to evaluate transportability of in-vitro cellular models, informing assay prioritisation and supporting novel hypothesis generation.

The extent to which in-vitro cellular experiments are effective models for complex biology underlying human phenotypes is often unclear. Here the authors propose and implement a framework using gene perturbation data to evaluate in-vitro models.