Representing high throughput expression profiles via perturbation barcodes reveals compound targets

Abstract

High throughput mRNA expression profiling can be used to characterize the response of cell culture models to perturbations such as pharmacologic modulators and genetic perturbations. As profiling campaigns expand in scope, it is important to homogenize, summarize, and analyze the resulting data in a manner that captures significant biological signals in spite of various noise sources such as batch effects and stochastic variation. We used the L1000 platform for large-scale profiling of 978 representative genes across thousands of compound treatments. Here, a method is described that uses deep learning techniques to convert the expression changes of the landmark genes into a perturbation barcode that reveals important features of the underlying data, performing better than the raw data in revealing important biological insights. The barcode captures compound structure and target information, and predicts a compound’s high throughput screening promiscuity, to a higher degree than the original data measurements, indicating that the approach uncovers underlying factors of the expression data that are otherwise entangled or masked by noise. Furthermore, we demonstrate that visualizations derived from the perturbation barcode can be used to more sensitively assign functions to unknown compounds through a guilt-by-association approach, which we use to predict and experimentally validate the activity of compounds on the MAPK pathway. The demonstrated application of deep metric learning to large-scale chemical genetics projects highlights the utility of this and related approaches to the extraction of insights and testable hypotheses from big, sometimes noisy data.

Deep Learning Benchmarks on L1000 Gene Expression Data

“…Other authors have used neural network models to build embeddings of gene expression data. In [14], the authors use a twin network architecture to represent gene expression profiles as 100 dimensional bar-codes. Comparing their representations to the standard representation (raw z-scores) and a representation based on gene set enrichment analysis (GSEA) they find that their perturbation barcodes consistently identify replicates, samples generated from perturbagens with shared targets, and…”

Applications of Deep-Learning in Exploiting Large-Scale and Heterogeneous Compound Data in Industrial Pharmaceutical Research

“…With one of these technologies, known as L1000™ Expression Profiling (profiling for 978 gene expressions) (De Wolf et al, 2016; ), thousands of compounds can be screened per day at lower costs than conventional microarray techniques (Subramanian et al, 2017). Merck reported the screening of a set of 3,699 compounds using the Genometry L1000 platform to unveil a new target for compounds (Filzen et al, 2017). Janssen announced (How library-scale gene-expression profiling is changing drug discovery; Pascale, 2015) that they will use Genometry’s L1000 platform to generate gene-expression profiles for 250,000 compounds from Janssen’s small-molecule screening library.…”

A Bayesian approach to accurate and robust signature detection on LINCS L1000 data

“…Among all query size we tested, our Bayesian method has higher TPR at the same FPR and the performance difference is the most significant when the query size is small. To better illustrate the performance of three methods under different query size, we show the median quantile of GSEA scores between bio replicates, 9 which is equivalent to the FPR at TPR = 0.5. As shown in Fig.…”