Benchling launches AlphaFold beta feature — seamlessly integrating deep learning into biomolecule design

Ashu Singhal

AlphaFold is an artificial intelligence program developed by DeepMind that can predict the 3D structure of a protein from an amino acid sequence with unprecedented accuracy. Not only is it a scientific breakthrough with massive potential, but it’s also emblematic of the new era of modern biotech — data-driven, open-sourced, collaborative and ultimately, faster than ever. The more we can accelerate our understanding of the building blocks of cells, the more we can speed up advanced drug discovery.

Unlocking AlphaFold’s potential

But how does biotech actualize AlphaFold? Despite all the hype, the vast majority of labs are unable to access AlphaFold today. AlphaFold is open source to use, but setting up the machine learning architecture to run the AlphaFold algorithm is extremely complicated, and takes significant engineering bandwidth to use in a stable, sustainable way. Researchers agree, it will take time and experimentation to determine how to best wield AlphaFold and related AI tools.

Enter Benchling

Our team gets excited about two things: science and bringing software to science. So it’s no surprise that when DeepMind made AlphaFold open source available for commercial use at the end of January 2022, our team jumped on the opportunity in a hackathon.

We focused on two core challenges that we were seeing in the R&D community with AlphaFold adoption. First, can we improve access to AlphaFold for the typical scientist with limited computing resources? And second, can we integrate AlphaFold into the Benchling platform in such a way that the structure predictions can be leveraged in actual scientific workflows?

Introducing Benchling AlphaFold beta feature

The Benchling AlphaFold beta feature allows customers to select any amino acid sequence stored in Benchling, request a 3D structure for it, and visualize the results in our platform. Customers can view and interact with the 3D structures in a Molstar (Mol*) viewer alongside the primary sequence. The structure files (.pdb format) also can be downloaded for more sophisticated modeling using third party applications. Scientists may readily share these protein structure files with other teammates, further extending the reach and utility of the data output.

Now for the first time, with our AlphaFold beta feature, scientists can not only predict 3D structures of novel proteins directly within Benchling, but also centralize experimental context, collaborate with teammates, and connect with downstream scientific workflows on a single, secure platform.

AlphaFold beta

The ease of access for scientists via Benchling is important because AlphaFold’s predictive AI is still not widely accessible to typical scientists today. By making it available at the click of a button, scientists will be able to try it out and find new ways to leverage AlphaFold output in their research. The current path to 3D structure resolution takes painstaking lab work over many months, even years. The cost and technical complexity is prohibitive, especially for resolving the structures of unknown proteins at scale.

A mindset shift

Tools like AlphaFold and neural network-based algorithms are no doubt quickly changing the R&D landscape. Over just the next few years, these tools can bring breakthroughs not yet conceivable. The power of AlphaFold also ushers in a mindset change, with biologists now more open to insights from computational approaches and adjusting to the new pace of modern R&D.

This massive shift we’re seeing away from an empirical approach to biology to an engineered biology where we can now read, write and edit the building blocks of life, is really just beginning. That means we need to keep investing in our product and incorporating new capabilities to match the innovation of our customers. For example, over the last year Benchling has developed functionality to solve new needs of our customers, such as support for chemically modified RNA, high-throughput lab automation, and native advanced analytics in platform.

While the use cases for AlphaFold are still being explored and proven, Benchling’s goal with our beta feature is to support our community in this truly exciting exploration. Early customer feedback has been positive — customers are looking to embrace AlphaFold and now see a seamless path to research via the beta feature.

AlphaFold signals how the world is redefining what science is and how it’s done. In this case, we see the combination of amazing machine learning (ML) expertise, domain expertise, and computational power. There are so many more data-rich research domains where ML can change how we do R&D, in climate science, consumer packaged goods (CPG), foods and more. At Benchling, we know that this is just the beginning.

NOTE: Benchling's beta feature for protein structure prediction using AlphaFold is not a generally available feature at this time. Current customers can sign-up for the waitlist in Benchling. Any unreleased services or features referenced in this blog post are still pending general availability.

Early beta users

PetMedix is developing therapeutic antibodies for companion animals, and having the ability to produce AlphaFold structures of our antibodies and antigens allows us to better understand the biology behind them. There has been a lot of technological developments in Artificial Intelligence that are now being applied to answering biological questions in important fields such as immunology. We see this in the literature, for example in the study of COVID19 immune response and antibody design, and we are excited to be able to apply these technologies to our antibodies, so we can help save and improve the lives of animals all over the world.
Dr. Albert VilellaHead of Bioinformatics at PetMedix

Sign up for the Benchling newsletter

Get our latest insights and announcements every month.

Powering breakthroughs for over 1,200 biotechnology companies, from startups to Fortune 500s

Helix Image