Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Tools of the Trade
  • Published:

An AI pipeline to investigate the binding properties of poorly annotated molecules

Nature Reviews Physics volume 4, page 359 (2022)Cite this article

Subjects

Binding between proteins and small molecules is an important link-prediction task affecting biological processes such as metabolic reactions, regulatory mechanisms and signal transduction pathways. However, binding annotations are available only for a small subset of the possible combinations of proteins and molecules. For instance, in the Foodome project, we have data for more than 135,000 small molecules found in foods, but only 4.58% of the collected molecules have known binding annotations with proteins. Machine learning (ML) models aim to fill the gap by predicting binding, but — given the paucity of data — their reliability remains questionable. Similar issues affect all areas leveraging sparse and biased network data such as recommendation systems, citation networks and collaboration networks.

We showed analytically that excellent performance in transductive tests can be achieved by algorithms that are much simpler than neural networks and that disregard the molecular structures of proteins and ligands. These configuration models are inspired by network science, and we use them as the baseline of the binding prediction task. Their success in transductive tests is driven by the underlying network topology of the training datasets, which presents an extremely biased picture of binding: molecular structures tend to either have an excessive ratio of positive annotations (binding) to negative (non-binding), or vice versa. In scenarios affected by this annotation imbalance, ML models behave like configuration models, disregarding structural information, and performing poorly in inductive tests.

This is a preview of subscription content, access via your institution

Relevant articles

Open Access articles citing this article.

Access options

Buy this article

  • Purchase on Springer Link
  • Instant access to full article PDF
Buy now

Prices may be subject to local taxes which are calculated during checkout

Acknowledgements

The author would like to acknowledge the whole team of experts behind AI-Bind and, in particular, A. Chatterjee, T. Eliassi-Rad and A.-L. Barábasi. The author also thanks C. Both and G. Kitzinger for helping in the visualization of the protein–ligand bipartite network.

Author information

Authors and Affiliations

  1. Network Science Institute and Department of Physics, Northeastern University, Boston, MA, USA

    Giulia Menichetti

  2. Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, USA

    Giulia Menichetti

Authors
  1. Giulia Menichetti
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Giulia Menichetti.

Ethics declarations

Competing interests

The author declares no competing interests.

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Menichetti, G. An AI pipeline to investigate the binding properties of poorly annotated molecules. Nat Rev Phys 4, 359 (2022). https://doi.org/10.1038/s42254-022-00471-1

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/s42254-022-00471-1

This article is cited by

Search

Advanced search

Quick links

Nature Briefing AI and Robotics

Sign up for the Nature Briefing: AI and Robotics newsletter — what matters in AI and robotics research, free to your inbox weekly.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing: AI and Robotics