mRNA and Artificial Intelligence: An Interview with Anima Biotech’s Yochi Slonim
In the life sciences industry, there is a wide variety of directions that companies can go. Many high-profile biopharma companies have a focus on messenger RNA (mRNA), and some are broadening their horizons by taking an interest in artificial intelligence. One company, Anima Biotech, has a unique approach to both.
Anima Biotech, headquartered in Bernardsville, NJ, is merging mRNA therapeutics with its own AI and bioinformatics capability. Yochi Slonim, co-founder and chief executive officer of Anima Biotech, took time to speak with BioSpace about the company and mRNA therapeutics.
Slonim started with an introduction to genetic translation, which is the process of mRNA delivering genetic code to ribosomes, and drug development. Slonim noted that traditional drug development involves small molecules. These small molecules, he said, “bind to the targets, a pocket in the protein, and disarm the protein from doing its chemical activity. When there is no lock, they are called hard targets or undruggable targets, because the industry for over 70 years has been unable to find them.”
The industry has been quite successful at finding low-hanging fruit, the actual druggable enzymes, because they have big pockets that are, Slonim said, “nicely shaped. But not all do and are very hard to reach, and many don’t have them at all.”
There have been approaches with nucleic acid drugs that bind to the mRNA and sort of kill it, called “knockdown.” Slonim observes that this is something of a one-way street. “It just knocks them down. Whenever diseases lack a protein or are underexpressed, we don’t know how to increase the protein, only how to kill it after it is made or prevent it from being made.”
A new approach is to try and knock down the mRNA itself with small molecules, but Slonim says, “mRNA is a very hard target itself, much harder than protein. It’s a very big molecule, and it’s not completely clear what these molecules can do or what will happen if you apply these small molecules to an mRNA or ribosome.”
This brings the subject around to Anima Biotech’s approach, which is in the mRNA small molecule space. Slonim says, “We have discovered molecules that control the translation of the mRNA by ribosomes into protein. We can actually discover molecules that do that specifically for individual proteins because we go after a new target, a rational target instead of the mRNA itself.”
And the technology uses actual imaging techniques, which they merge with an AI system that scans and analyzes the images. “Our technology enables us to visualize, literally, visualize with light, the production of proteins by ribosomes,” Slonim says.
The core technology of the company was developed at the University of Pennsylvania and it labels the transfer RNA (tRNA), which acts as the physical link between the mRNA and the amino acid sequence of proteins. The patented technology for this leverages FRET signals (Fluorescence Resonance Energy Transfer). FRET basically means that when two molecules come close together and if they are labeled with fluorescent dyes, there is an energy transfer. And that energy transfer in labeled molecules causes an increase in light—the more energy transfer, the more light, meaning the more protein production, the more light.
Slonim says, “Ribosomes pair together amino acids, brought together by tRNA molecules, and this occurs millions of times over and over again in each of the trillions of cells in the body—it’s all they do, pair together amino acids to build the chain of the protein.”
What Anima’s technology platform allows them to do is select a signature pair for each target protein and label them. “So the light you see is highly correlated to the target protein,” Slonim says. “It’s the proteins being assembled by ribosome in real-time. And you can use it as a primary assay as a screening system that scans molecules. The molecules that create light means it is creating the production of the target protein. Increasing the light means it’s increasing the target protein.”
That means the company’s technology can screen a library of compounds and actually identify if a compound is interacting with a ribosome and whether it is increasing protein production or decreasing protein production. “Our technology enables us to do this for almost every disease, potentially. And the technology of the company has evolved over the years into a product. We are now building a pipeline that currently includes multiple programs,” Slonim says.
For most of the programs they are in the “hit to lead” area, prior to a preclinical and clinical program, although Slonim believes they will be into the clinic in 18 months to two years. Currently, their programs are for Collagen I for lung fibrosis, liver fibrosis and scleroderma, RSV for viral translation inhibitors, C-Myc for oncology, and Huntingtin for Huntington’s disease.
The company also has a major partnership deal with Eli Lilly. Although he won’t discuss the specific targets or diseases, they are utilizing their technology platform on targets selected by Lilly.
How does the AI fit in? The light created by the images is too subtle for the human eye to evaluate effectively, and there’s just too much data. For example, a typical compound library contains 200,000 compounds. Slonim notes that in a typical project like the one with Lilly, there will be multiple cells that act as samples of the disease and in each one there will be a different compound out of the 200,000 compounds. And an automated system runs them under the microscope.
They may get 5,000 individual images multiplied by 200,000 compounds, with each light representing production of the protein. That creates a billion examples, and since they take images on several channels, there’s something like 5 billion examples and hundreds of data points, which creates 5 to 7 billion data points.
“We cannot look at those images one by one,” Slonim says. “We can’t hire 2,000 people in India to look at them, and even if you could, the increase in light might be by 20% and that light could represent a possible drug, your eyes couldn’t see the difference.”
But the AI can. “We use AI to actually identify the activity, the molecules that are actually the ones doing the magic. It’s like finding maybe 20 out of 200,000 but looking at a billion data points in order to do that.”
Their AI system also learns and Slonim compares it to facial recognition software. “It works differently, but the outcome is that with this technology we are teaching the system what it can see in the millions of examples of compounds that don’t work and the ones that do. With each screening with a new target, the system learns more and more and is more able to discover new drugs. The more we discover the more it CAN discover.”
And Slonim feels that it’s this combination of good biological understanding and cutting-edge AI software that gives Anima Biotech an edge in this field.
Asked about how biopharma is generally using AI, Slonim compares it to a horse and carriage. “I think there are two types of companies. Most in the field of AI as applied to drug discovery have a dilemma. There is a carriage but no horse or a horse but no carriage to carry. Taken to an extreme example, there is a concept where you don’t take any particular biological data, you just take cells and flow compounds into them. But rather than look at this particular data, such as the light coming from ribosomes, you have the computer take images and it learns by itself what is healthy and what is sick. Out of these millions of compounds, the computer is guessing if the cells are better than before. It’s high-scale screening, but no biology.”
Slonim and Anima Biotech believe that their advantage is that they also have a novel biological mechanism to investigate as well as an imaging system and AI analysis system.
He also adds that, “business-wise, if you have this idea that you have compounds but no deep understanding of how they work and what they do, it’s very hard for pharma companies to partner with you.”
As a result, many AI companies partnering with pharma companies do so as a service model, rather than as a drug discovery partnership. “I think it’s better to start with the biology, generate new biological data to give you insight about something that happened in the disease, and use the computer to provide understanding and meaning to what you’re seeing. That’s the power of the computer, to analyze huge amounts of data, not to get what is a disease and what is not. I think biologists should be working together with the AI guys, side by side—that’s the winning approach,”