The ultra-high processing power of deep machine learning is fuelling researchers’ ability to analyse data, automate laboratory systems and conduct novel synthetic biological experiments.
It seems unlikely, but there is a direct link between the US Department of Defense (DoD), a young Australian firm specialising in fermentation, and synthetic biology (synbio) and deep machine learning (DML) technologies.
The DoD predicts the next industrial revolution will be a “biomanufacturing revolution”, according to Heidi Shyu, Under Secretary of Defense for Research and Engineering.
To get ready, Shyu said the department is boosting biomanufacturing capabilities in five areas critical to national security: fabrication, firepower, fitness, food and fuel.
To that end, the DoD awarded US$1.76 million last November to Cauldron Molecules. Its job is to plan a commercial-scale facility for creating “precision-fermented products”, ranging from foods and fuels to chemicals.
Enter synbio, which combines engineering principles and advanced genetic technology to rapidly design, test and build novel cell-based parts, products and systems.
And DML provides the predictive power driving complex synbio outcomes. This is why Cauldron is seeking to better understand DML, said CEO and Co-Founder Michele Stansfield.
From AI to DML
At its simplest, DML is a form of artificial intelligence (AI). From this basis, machine learning can perform more complex tasks without explicit instructions. It can, to illustrate, quickly harvest data, identify patterns and make predictions.
Machine learning is applied across a broad range of areas and it does so with minimal human interference. It is used, for example, in finance to make trading decisions. In energy production and distribution, machine learning is the core of so-called “smart grids”. In health care, it is used to help diagnose diseases and apply epidemiological findings.
And in media and entertainment, systems powered by machine learning can make content recommendations.
But DML cranks-up the complexity of jobs a machine can accomplish and without human intervention. It does so by mimicking the human brain. Data passes through a web of interconnected machine neurons, or algorithms, much as information is transmitted and processed in the human brain.
There are three types of these brain-like networks:
- Recurrent neural networks remember past events and make predictions about the future.
- Convolutional neural networks are more complex. They split data into small segments and make conclusions based on analysis of the results, and are ideal for applications such as image processing.
- The third and newest neural network is the transformer, which assesses multiple sources of data simultaneously, and is widely used in tasks associated with language.
By offering these types of networking, DML is the form of AI that most closely mimics the human brain. DML can therefore outperform traditional machine learning in complex pattern recognition tasks such as image classification, object detection, natural language processing, and data filtering and recommendation.

Today and tomorrow
DML is already hard at work. For example, the chatbot ChatGPT is based on a transformer architecture. “GPT” stands for generative pre-trained transformer.
Other everyday applications of DML reflect its emergence as a critical component of current and future technologies. It is already the computer engine powering digital assistants, automatic facial recognition, fraud detection and even the voice-activated TV remote found in households nationwide.
It is no surprise that Elon Musk is hiring “deep learning” engineers for Tesla’s robotic humanoid project Optimus. After all, DML is central to self-driving cars. They use it now to detect road signs and pedestrians.
DML is also used by experts to identify areas of interest in defence system satellite images. Radiologists look for signs of cancer with the assistance of DML-based medical imaging applications. And in a change of pace, DML was the engine that enabled a computer program, AlphaGo, to beat a human player in the 2015 DeepMind Challenge Match.
In sum, the benefits of DML over AI and machine learning are numerous: from the efficient processing of unstructured data and the ability to tease out hidden relationships and patterns, to the ability to process the large variations in volatile datasets, all without human participation.
The synbio link
As William Beardall and his colleagues at Imperial College London and Boston University noted, because of its human-like processing system, DML has a “natural synergy” with synbio.
DML is already boosting the power of synbio to drive advances in many current research areas.
Among these are the design and simulation of novel biological components, automated analysis of imaging data, structure-based learning, not to mention algorithms for interpreting protein sequences in microbes and simple organisms such as yeast. Researchers also use DML to design and automate laboratory experiments.
“Overall, deep learning methods have already had a substantial impact on the field of synthetic biology,” the team wrote in Genetic Engineering & Biotechnology News. “We anticipate significant advances in this area moving forward.”