The pharmaceutical industry has pivoted from manufacturing small molecules to biologics in less than a decade, and now is bringing forward new technologies to adapt. That’s the view of Claes Gustafsson, PhD, co-founder and CCO of ATUM, who will speak at the Bioprocessing Summit Europe in March 2024 about innovation to maximize monoclonal antibody production.
According to Gustafsson, new tools such as artificial intelligence (AI), when combined with synthetic biology (synbio), can work together to accelerate antibody manufacturing by improving the manufacturability of new drugs.
“Fundamentally, with machine learning and AI, you can design any sequence you want in a computer, and synbio allows you to build it to test in a biological system,” he says.
As Gustafsson explains, the pharma industry for a century was driven by small molecule drugs—which are largely manufactured using chemical processes. However, all this changed with the invention of biological therapies, such as monoclonal antibodies and mRNA.
While in 2010 nine out of the top ten drugs by revenue were small molecules, just two years later (in 2012), seven were biologics, Gustafsson says.
“One of the big problems is making them,” he points out, explaining that inorganic chemical processes are more clear cut than biological manufacturing, which can be expensive, complex, and unpredictable.
Maximizing biologics manufacturing
To overcome these issues, he says that ATUM and the wider industry are turning to innovative technologies to maximize the manufacturing of biologics.
“What drives us, and the entire industry, is how to forefront technology to engineer biology in a faster, more efficient, and more predictable fashion,” he continues.
Among these new technologies is the use of large language models (LLMs) to analyze databases of antibodies looking for patterns of “genetic language.” Such analyses can be used to improve antibody stability or manufacturing yields.
They can also be used to open new design space for antibodies by, for example, looking up historical genetic sequences.
“If you think about machine learning as an LLM, you’re going to find all existing sequences, but you can use the same information to calculate ancestral nodes that ceased to exist on Earth 2–3 billion years ago,” notes Gustafsson, adding that this type of analysis can help design antibodies with better binding to drug targets and reduced off-target effects.
Synbio, meanwhile, can be used to rapidly build new genetic elements that can be tested in biological systems to see, for example, whether they improve protein expression during cell-line development. Another technology involves “jumping genes” (transposons), which Gustafsson says Atum has built using machine learning and synbio to more efficiently build cell lines that produce, for example, monoclonal antibodies at a higher yield.
“AI and machine learning on their own are useless, unless you tie in the design process with physical implementation, testing designs within biological systems as you move toward a commercial endpoint,” he explains.