Active machine learning-driven experimentation to determine compound effects on protein patterns

  1. Armaghan W Naik
  2. Joshua D Kangas
  3. Devin P Sullivan
  4. Robert F Murphy  Is a corresponding author
  1. Carnegie Mellon University, United States

Abstract

High throughput screening determines the effects of many conditions on a given biological target. Currently, to estimate the effects of those conditions on other targets requires either strong modeling assumptions (e.g. similarities among targets) or separate screens. Ideally, data-driven experimentation could be used to learn accurate models for many conditions and targets without doing all possible experiments. We have previously described an active machine learning algorithm that can iteratively choose small sets of experiments to learn models of multiple effects. We now show that, with no prior knowledge and with liquid handling robotics and automated microscopy under its control, this learner accurately learned the effects of 48 chemical compounds on the subcellular localization of 48 proteins while performing only 29% of all possible experiments. The results represent the first practical demonstration of the utility of active learning-driven biological experimentation in which the set of possible phenotypes is unknown in advance.

Article and author information

Author details

  1. Armaghan W Naik

    Computational Biology Department, Center for Bioimage Informatics, Carnegie Mellon University, Pittsburgh, United States
    Competing interests
    The authors declare that no competing interests exist.
  2. Joshua D Kangas

    Computational Biology Department, Center for Bioimage Informatics, Carnegie Mellon University, Pittsburgh, United States
    Competing interests
    The authors declare that no competing interests exist.
  3. Devin P Sullivan

    Computational Biology Department, Center for Bioimage Informatics, Carnegie Mellon University, Pittsburgh, United States
    Competing interests
    The authors declare that no competing interests exist.
  4. Robert F Murphy

    Computational Biology Department, Center for Bioimage Informatics, Carnegie Mellon University, Pittsburgh, United States
    For correspondence
    murphy@cmu.edu
    Competing interests
    The authors declare that no competing interests exist.

Reviewing Editor

  1. Uwe Ohler, Duke, Germany

Version history

  1. Received: July 13, 2015
  2. Accepted: January 28, 2016
  3. Accepted Manuscript published: February 3, 2016 (version 1)
  4. Version of Record published: March 7, 2016 (version 2)

Copyright

© 2016, Naik et al.

This article is distributed under the terms of the Creative Commons Attribution License permitting unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 6,881
    Page views
  • 1,213
    Downloads
  • 34
    Citations

Article citation count generated by polling the highest count across the following sources: Scopus, Crossref, PubMed Central.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Armaghan W Naik
  2. Joshua D Kangas
  3. Devin P Sullivan
  4. Robert F Murphy
(2016)
Active machine learning-driven experimentation to determine compound effects on protein patterns
eLife 5:e10047.
https://doi.org/10.7554/eLife.10047

Share this article

https://doi.org/10.7554/eLife.10047

Further reading

    1. Cell Biology
    Kazuki Hanaoka, Kensuke Nishikawa ... Kouichi Funato
    Research Article

    Membrane contact sites (MCSs) are junctures that perform important roles including coordinating lipid metabolism. Previous studies have indicated that vacuolar fission/fusion processes are coupled with modifications in the membrane lipid composition. However, it has been still unclear whether MCS-mediated lipid metabolism controls the vacuolar morphology. Here, we report that deletion of tricalbins (Tcb1, Tcb2, and Tcb3), tethering proteins at endoplasmic reticulum (ER)–plasma membrane (PM) and ER–Golgi contact sites, alters fusion/fission dynamics and causes vacuolar fragmentation in the yeast Saccharomyces cerevisiae. In addition, we show that the sphingolipid precursor phytosphingosine (PHS) accumulates in tricalbin-deleted cells, triggering the vacuolar division. Detachment of the nucleus–vacuole junction (NVJ), an important contact site between the vacuole and the perinuclear ER, restored vacuolar morphology in both cells subjected to high exogenous PHS and Tcb3-deleted cells, supporting that PHS transport across the NVJ induces vacuole division. Thus, our results suggest that vacuolar morphology is maintained by MCSs through the metabolism of sphingolipids.

    1. Cell Biology
    2. Chromosomes and Gene Expression
    Monica Salinas-Pena, Elena Rebollo, Albert Jordan
    Research Article

    Histone H1 participates in chromatin condensation and regulates nuclear processes. Human somatic cells may contain up to seven histone H1 variants, although their functional heterogeneity is not fully understood. Here, we have profiled the differential nuclear distribution of the somatic H1 repertoire in human cells through imaging techniques including super-resolution microscopy. H1 variants exhibit characteristic distribution patterns in both interphase and mitosis. H1.2, H1.3, and H1.5 are universally enriched at the nuclear periphery in all cell lines analyzed and co-localize with compacted DNA. H1.0 shows a less pronounced peripheral localization, with apparent variability among different cell lines. On the other hand, H1.4 and H1X are distributed throughout the nucleus, being H1X universally enriched in high-GC regions and abundant in the nucleoli. Interestingly, H1.4 and H1.0 show a more peripheral distribution in cell lines lacking H1.3 and H1.5. The differential distribution patterns of H1 suggest specific functionalities in organizing lamina-associated domains or nucleolar activity, which is further supported by a distinct response of H1X or phosphorylated H1.4 to the inhibition of ribosomal DNA transcription. Moreover, H1 variants depletion affects chromatin structure in a variant-specific manner. Concretely, H1.2 knock-down, either alone or combined, triggers a global chromatin decompaction. Overall, imaging has allowed us to distinguish H1 variants distribution beyond the segregation in two groups denoted by previous ChIP-Seq determinations. Our results support H1 variants heterogeneity and suggest that variant-specific functionality can be shared between different cell types.