Gibson SOLA Platform

Generative AI is changing how scientists design new medicines. Traditional drug discovery can be slow and expensive, especially when it comes to finding the right molecule to fight a specific disease. With the help of powerful computer models, scientists can now create and test molecular designs much faster than ever before. Two of the biggest breakthroughs in this area are generative adversarial networks (GANs) and diffusion models. These systems can design completely new molecules—sometimes so new they wouldn’t have occurred to any human researcher.

GANs use a “generator” that attempts to create something such as a new molecule while a “discriminator” judges if that molecule matches desired criteria. These two networks learn from each other, so the generator keeps improving its designs until the discriminator can’t tell them apart from a known set of good examples (figure 1). GAN-based tools such as the DrugGEN system, which has been used to identify AKT1 inhibitors for the treatment of certain types of cancer, and MedGAN which can predict effective molecular structures to aid in repurposing existing drug candidates, are redefining how traditional drug discovery can be accomplished.

Meanwhile, diffusion models start with random noise and repeatedly remove that noise in tiny steps. Eventually, they form a clear image of a molecule that could meet the specified criteria. The ability of these models to incorporate structure-based conditioning, multi-modal inputs, and gradient-free sampling makes them extremely powerful for next-generation drug discovery pipelines. With both methods, what used to take years of trial and error in a lab can now happen digitally in weeks or even days.

Figure 1. GAN Networks. Generative Adversarial Networks (GANs) use a generator to create novel molecular candidates. These are evaluated by a discriminator against known molecules, providing feedback that improves future outputs. Gibson SOLA Platform is a ‘digital to biological converter’ that generates DNA or mRNA in less than one day.

Antibodies are among the most exciting targets for these AI systems. Antibodies are proteins the body uses to fight infections, and they’re also the basis of many modern therapies. To train an AI model to invent an antibody from scratch, scientists feed it huge amounts of data on existing antibodies, along with details on how they bind to specific targets. The AI models can suggest unique designs that are predicted to bind more efficiently than known antibodies and can even improve the generation of engineered multivalent antibodies to further improve avidity and efficacy. Of course, once the AI model generates these promising ideas, researchers must still test them in cells or animals. But researchers can skip a lot of the guesswork, because they already have reason to believe these designs will do well.

Another fascinating area is the use of AI for mRNA-based therapies. Over the past few years, mRNA has gained attention for its utility in vaccines and other treatments. Designing mRNA that’s stable and effective isn’t simple. The RNA sequence can affect how well it’s translated into proteins and how the body’s immune system responds to it. With AI models, researchers can explore thousands of possible mRNA sequences and quickly narrow down which ones are most likely to succeed. They can then bring those sequences into the lab to evaluate performance and efficiency.

Figure 2. Complete projects more quickly. Eliminate bottlenecks and reduce dependence on external providers by building biology overnight. Historically, service providers can require up to 6 weeks to prepare transfection-grade plasmids (top). The Gibson SOLA Platform generates pDNA in 1 day (bottom).

A critical complement to AI models is the ability to generate DNA and mRNA on-demand, even overnight for fastest results. Historically, sequence designs were sent to an external service provider who would need a month or longer to generate the DNA or mRNA. If the sequences fail upon arrival, the process begins again with a new timeline, extending the time to answers. Companies that have their own synthesis capabilities can produce the DNA or mRNA in-house quickly. They proceed from digital sequences to biological data in a much shorter time frame which allows for significantly faster iterative cycles. These companies also retain confidentiality of the designs by preventing them from being added to databases maintained by service providers.

Telesis Bio’s novel Gibson SOLA Platform empowers researchers to make the DNA or mRNA they need, when they need it. The platform assembles DNA from universal building blocks on-demand, without the hassle that usually comes with building large or complex sequences. It can also be adjusted for mRNA-related projects, which is a critical advancement for researchers working on vaccines or gene therapies. Instead of waiting for outsourced DNA or mRNA, researchers can analyze lab results within days. The Gibson SOLA Platform is a game changer that removes the barriers and delays that were previously considered unavoidable (figure 2).

Consider a realistic scenario in which a proprietary AI model designs a new antibody that targets a cancer-related protein more effectively than anything else known. Using the Gibson SOLA Platform, that team assembles the template DNA for that antibody on-demand. The DNA is transfected into cells, producing a small batch of the antibody, which is tested for reactivity with cancer cells. If the results look good, the team can move on to more complex studies. If the results are disappointing, the designs are iteratively improved until a candidate is selected. Teams that generate DNA or mRNA in-house using the Gibson SOLA Platform achieve research milestones much faster than teams who outsource DNA or mRNA generation to external service providers (figure 3).

Figure 3. Automated DNA synthesis platform in your lab. The Gibson SOLA Platform is a ‘digital to biological converter’ that generates DNA and RNA on-demand in your lab. Maintain your sequences as trade secret by generating DNA in-house rather than sending them out to an external service provider for inclusion in their databases.

In the future, these techniques could lead to personalized medicine, where AI algorithms generate treatments tailored to an individual’s genetic makeup or their specific version of a disease. And because in-house synthesis and testing processes are now more accessible, labs of all sizes could potentially create breakthroughs in medicine, food security, and protein engineering.