CHICAGO – This month, DNAStar formally launched NovaFold AI, an update to its NovaFold protein structure prediction software that incorporates the AlphaFold 2 artificial intelligence system from Google sister company DeepMind Technologies. The company had introduced a beta version of the product, dubbed NovaFold AI powered by AlphaFold 2, in February.
Steve Darnell, team leader of DNAStar's structural biology and protein team, called NovaFold AI a cloud version of the AlphaFold 2 pipeline. The offering lets users access the AlphaFold protein structure prediction method from within DNAStar's NovaCloud Services interface of the firm's Protean 3D visualization and analysis suite.
He said that AlphaFold 2 is designed to help researchers without a bioinformatics background predict structures.
"This release allows customers to more tightly integrate the analysis with running the prediction," added DNAStar General Manager Shawn Grass.
Madison, Wisconsin-based DNAStar sees NovaFold AI as a growth opportunity.
"The structure market has been a little more challenging for us to get into, because when people think of us, they think of just DNA. They don't think of us in the protein area," Grass said. "Our hope is with the excitement and buzz right now about structure prediction that this will give us an opportunity to expand our reach."
DNAStar, which has been in business since 1984, is actually quite active beyond DNA.
Last year, the privately held firm introduced version 17.3 of its Lasergene software for DNA, RNA, and protein sequencing assembly and analysis. That release added viral genome analysis workflow support for Pacific Biosciences and Oxford Nanopore long-read sequences, including data from PCR-amplified fragments generated according to Artic network protocols. The update was also optimized for genomic analysis and variant identification to support COVID-19 research.
Lasergene, which serves the molecular biology and genomics markets, is one DNAStar's two product lines. It includes a series of applications for tasks like genome visualization, sequence assembly, sequence alignment, and in silico cloning. DNAStar's molecular visualization and analysis platform, Protean 3D, is technically part of Lasergene as well.
"Protean 3D has a tight integration between its sequence representations and its structural representations through a selection model," Darnell explained.
Lasergene products are "general tools" aimed at anyone in molecular biology or genomics looking to manage pipelines and workflows, according to Grass. He said that a fair number of customers mistakenly refer to the company as Lasergene because the product name is well known in some circles.
The other DNAStar product line, for structural biology applications including structure prediction, docking, and antibody modeling, is called Nova.
NovaFold is the commercialization of I-TASSER, protein structure prediction software developed by University of Michigan bioinformatician Yang Zhang. In 2015, DNAStar licensed the exclusive commercial rights to I-TASSER and markets the technology as part of Protean 3D.
The firm also licenses technology for its NovaDock protein-protein docking software from Cancer Research UK. Darnell said that the company's products represent a mixture of DNAStar-developed tools, freely available external tools like AlphaSeek, and licensed technology.
DNAStar is not directly partnering with DeepMind, which makes some of its AlphaFold technology available through open-source channels.
Grass said that DNAStar has not made any additions or other changes to the AI that AlphaFold provides. "It's something we are always looking at, but right now, we felt it was best to leave it just how it was," he said.
What the company does offer with the new release is the ability to add AlphaFold structure predictions as templates to the core NovaFold software to guide the modeling process toward these predictions. "What we also bring to the table is the additional downstream analysis outside of just coordinate generation," according to Darnell.
Most of the tools in both product lines are integrated on DNAStar's NovaCloud infrastructure, which the firm has been building up for the last five or six years.
DeepMind's AlphaFold had the highest score in the 14th and most recent Critical Assessment of Structure Prediction (CASP) competition in 2020, in which entrants are given the amino acid sequences for about 100 proteins to then predict their structures.
"It certainly excels wonderfully at fold recognition and modeling," Darnell said. He added that AlphaFold will be entered in CASP 15 this summer, a competition that will have a greater focus on predicting multimers.
That is a direction DNAStar wants to go in with NovaFold 2.
"The direct structure prediction of multimers is definitely where I believe the field is certainly looking forward toward, as well as different and better approaches for doing multi-domain proteins," Darnell said. "A tool like AlphaFold … can identify and actually model different domains in isolation within a multi-domain sequence."
Darnell also expects to see more activity around protein-protein docking technology in subsequent years.