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.

Volume 559 Issue 7714, 19 July 2018

Parsed judgement

Whether it is bacteria following chemical gradients or humans unpicking a complex odour, the ability to recognize molecular patterns is a key attribute of organisms. DNA-based neural networks can perform the same task in molecular computing but have so far been limited to a set of no more than 4 patterns of 4 bits each. In this week’s issue, Lulu Qian and Kevin Cherry increase those numbers to 9 and 100, demonstrating a set-up that can correctly identify the numbers 1–9 written to a 10 × 10 pixel grid. Their network uses the ‘winner take all’ competitive strategy to refine its output and determine which of the numbers it is viewing. The researchers show that even when 30 of the 100 bits of a given pattern are flipped — analogous to the variation in handwritten characters — their network can ‘remember’ the pattern correctly and recognize the digit. This suggests that molecular computation circuits can classify highly complex and noisy information based on something akin to a memory.

Cover image: Olivier Wyart

This Week

News in Focus

Comment

Careers

Futures

  • Futures |

    Cold comfort.

    • Kurt Pankau

Research

Amendments & Corrections

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

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

Search

Quick links