Forward-looking: Researchers around the world are embracing DNA-based storage right now. Mixing digital data and biology could bridge the best of both worlds, though a few challenges are still slowing market and industry adoption.

Visionary solutions using DNA sequencing have been hailed as the future of the storage world for a few years now. Biology seems to have solved the data encoding problem a few billion years ago, so we could learn a thing or two from nature while we prepare to expand the world’s digital realm to 180 zettabytes – amounting to 180 billion terabytes – by the end of 2025.

Israeli researchers say they have found a way to significantly improve the data retrieval process, which is one of the biggest issues DNA storage technology is facing right now. A team at Technion – Israel Institute of Technology used a specifically trained AI model to speed up data recovery from DNA strands by 3,200 times. Needless to say, the process is still much slower than “modern” storage technologies available on the market.

The AI tech in question is known as DNAformer, and is based on a transformer model trained by Technion researchers on synthetic data. The data simulator that fed DNAformer was also created at Technion. The model can reconstruct accurate DNA sequences from error-prone copies and can boost data integrity even further thanks to a custom error-correcting algorithm designed to work well with DNA.

DNAformer is much faster at retrieving data than previously unveiled methods. The AI model can read 100 megabytes 3,200 times faster than the most accurate existing method, and can seemingly do so with no loss of data. Accuracy is improved by “up to” 40 percent as well, which can further decrease the total retrieval process time.

The Israeli researchers tested DNAformer’s capabilities on a tiny 3.1-megabyte data set, which included a color still image, a 24-second audio clip, a written piece about DNA storage, and some random data. The latter was useful to show how the model can behave when dealing with encrypted or even compressed digital data. The team achieved a “data rate” of 1.6 bits per (DNA) base in a high-noise regime, the official study says, cutting the time needed to read the data back from several days to just 10 minutes.

The Technion team said DNAformer will be further developed and tailored to different data storage needs. The technology can easily scale and adapt to various scenarios, with promising prospects for its adaptability. The researchers are already thinking about “market demands” and future improvements in DNA sequencing to improve their AI technology.