Artificial Neural Network made of DNA can identify handwritten numbers

A group of researchers at Caltech have created an artificial neural network comprised of DNA that can correctly identify handwritten numbers. This new development seems to show that artificial intelligence can be programmed into biomolecular circuits. This development was created by Lulu Qian, assistant professor of bioengineering and his team, and published in Nature.

Artificial neural networks consist of mathematical models that are inspired by the human brain, but a bit more simplified. They operate much like neuron networks do and can process complex information. The Qian Lab’s primary goal in this project was to program intelligent behaviours like computing, and decision making, using artificial networks that were made out of DNA.

“Though scientists have only just begun to explore creating artificial intelligence in molecular machines, its potential is already undeniable,” says Qian. “Similar to how electronic computers and smart phones have made humans more capable than a hundred years ago, artificial molecular machines could make all things made of molecules, perhaps including even paint and bandages, more capable and more responsive to the environment in the hundred years to come.”

Qian continues explaining his work by adding, “Humans each have over 80 billion neurons in the brain, with which they make highly sophisticated decisions. Smaller animals such as roundworms can make simpler decisions using just a few hundred neurons. In this work, we have designed and created biochemical circuits that function like a small network of neurons to classify molecular information substantially more complex than previously possible.”

One of the more difficult tasks in building artificial intelligence is in programming them to recognize human handwriting, which varies from person to person. This has been a huge challenge for developing electronic artificial neural networks in the past. One that Kevin Cherry, a student of Qian, was tasked with bypassing. Teaching an artificial intelligence to recognize human handwriting required Cherry to teach the AI to not only account for variations in handwriting, but to then compare it to an unknown number in their “memories” and decide what number it was.

Cherry, the first author of the paper, described to Nature how the team used designed DNA sequences to carry out the prescribed chemical reactions that would allow the AI to identify molecular handwriting accurately. The difference between molecular handwriting and visual handwriting is that visual handwriting varies in shape, whereas molecular handwriting doesn’t. This lack of geometry isn’t uncommon in natural molecular signatures but it still requires sophisticated neural networks to identify them.

The Process

Cherry built a DNA neural network that was able to differentiate between handwritten 6’s and 7’s and then tested 36 handwritten numbers and the test tube neural network’s ability to correctly identify them, which it did. His system, in theory, can classify over 12,000 handwritten forms of 6’s and 7’s. About 90 percent of those numbers that were taken from a database were classified into two possibilities. The developers chose to implement a “winner take all” strategy using DNA molecules that Cherry and Qian developed. The strategy used certain DNA molecules they called “the annihilator” to select a winner when determining the identity of an unknown number.

“The annihilator forms a complex with one molecule from one competitor and one molecule from a different competitor and reacts to form inert, unreactive species,” says Cherry. “The annihilator quickly eats up all of the competitor molecules until only a single competitor species remains. The winning competitor is then restored to a high concentration and produces a fluorescent signal indicating the networks’ decision.”

The team then built on the principles of the first DNA neural network to make it even more complex. Their goal is to develop even more complex neural networks that can learn and create memories from the examples added to the test tube. The hope is to train the AI to perform other tasks as well.


Featured Image Source:/ Olivier Wyart —Conceptual illustration of a droplet containing an artificial neural network made of DNA that has been designed to recognize complex and noisy molecular information, represented as ‘molecular handwriting.’

Story Source: https://www.sciencedaily.com/releases/2018/07/180704135320.htm

Journal Reference:

  1. Kevin M. Cherry, Lulu Qian. Scaling up molecular pattern recognition with DNA-based winner-take-all neural networksNature, 2018; DOI: 1038/s41586-018-0289-6

 

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.