In what is a biological breakthrough, researchers from the California Institute of Technology have constructed a test tube artificial neural network that can recognize ‘molecular handwriting’. The development overcomes a major problem with machine learning; that is the ability of a machine to correctly identify handwritten numbers.
Artificial neural networks are computing systems based on the biological neural networks that constitute animal brains. These systems can “learn” to perform tasks by examining examples and instead of being programmed with any task-specific rules. Neural networks consist of input and output layers together with a layer made up of units that can transform the input into something that the output layer can use. The primary use is to find or to interpret patterns which are far too complex for a human to resolve.
Additionally the research is a pioneering breakthrough in programming artificial intelligence into synthetic biomolecular circuits. According to the lead researcher, Professor Lulu Qian: “Though scientists have only just begun to explore creating artificial intelligence in molecular machines, its potential is already undeniable.”
In term of what these potentials might be, future artificial molecular machines could enable anything constructed from molecules, from paint to bandages, more capable and more responsive to their environment. With the new study, the focus was on handwriting. Human handwriting varies considerably and while a person can (often) scrutinize a scribbled sequence of numbers and interpret them; for machines this creates a challenging computational task.
The researchers showed how a neural network made out of specially designed DNA sequences can perform prescribed chemical reactions to accurately in order to identify “molecular handwriting.” Here each each molecular number was made up of 20 DNA strands selected from 100 molecules.
The research could become the basis for conducting medical diagnostics, such as detecting the presence of biomolecules, such as cholesterol or blood glucose. The new study has been published in the science journal Nature. The research paper is titled “Scaling up molecular pattern recognition with DNA-based winner-take-all neural networks.”
