While many AI companies focus on analyzing existing scientific literature, this approach faces a fundamental challenge: the reliability of the source material itself. “Up to 70% of experiments described in publications cannot be replicated,” notes Yochi Slonim, CEO of Anima Biotech. “You do it once and you publish it. You do it again, it doesn’t happen.”
The replication crisis
Slonim is referencing a seminal 2016 Nature survey that found that more than seven out of ten researchers had attempted and failed to reproduce another scientist’s experiments, and more than half had failed to reproduce their own experiments. The problem persists to the present day, prompting new initiatives aimed at transparency, such as open data sharing and preregistration of study designs.
While Anima has contributed to the scientific literature itself, with a growing list of peer-reviewed publications on protein synthesis monitoring and tRNA visualization, Slonim argues that relying solely on mining existing research isn’t enough.
Interrogating cellular disease activity with AI
Yochi Slonim
Like many startups in the pharma space, Anima Biotech focuses on AI but isn’t applying traditional models to old processes or extracting findings from decades-old literature. “Let’s say you have an experiment showing that A affects B where AI and B are proteins,” Slonim explained. “A researcher might then find another study that finds B affects C, and another database where D is connected to C, creating a maze of connections.”
With its Lightning.AI platform, Anima is taking a “direct interrogation” approach. The company runs millions of parallel experiments examining cellular pathways in both healthy and diseased cells, generating roughly 100 million images in a typical run. Their system tests 500 different biological pathway hypotheses simultaneously, with each hypothesis tested a million times. Neural networks analyze these images in real-time, identifying and ranking pathways that show the most significant differences between healthy and diseased states. “There is a disease mechanism; there is something unique that you need to drug to correct,” notes Slonim.
“If aliens were coming with a spaceship landing on Earth and they want to drug diseases, they would bring on the spaceship a lab like this. They would laugh at how we are doing it. They’d say, ‘No, no, no, that’s not the way. We are going to just do—give us the disease. We are going to do a million experiments over 500 visualizers… and in a month, we’ll tell you.'”
Big data, tiny clues
Anima adds thousands of their own experimental results and 2 billion images that they generated over a decade. When identifying a potential disease mechanism, the company visualizes biological processes directly within living cells. “What people have been visualizing is cellular morphology, meaning they see the cells,” Slonim explains. “But there will be no hint as to what’s the biology behind it. The language of imaging doesn’t mean that we are talking biology.”
To decode that language, Anima has built an mRNA biology knowledge graph—a structured representation of data that contextualizes mRNA biology, including experimental results from billions of cellular images. “What we are doing is experimental biology at scale that is actually asking the disease itself, not the researchers who worked on the disease.”
“We are conducting a million experiments over 500 processes. A typical run like this will generate 100 million images. It goes to the cloud, and in the cloud, the neural network is actually looking at them. This happens in real time.”
From academia to industry
“The assets that we bring in are our unique neural networks trained on 2 billion mRNA pathway visualizations, hundreds of pathway visualizers, the world’s largest curated mRNA knowledge graph, and a specialized library of RNA-binding protein modulators.”
The approach has attracted significant pharmaceutical interest. Academic collaborators include universities like Duke, the University of Pennsylvania, the University of Oxford and research organizations such as the Scripps Research Institute. Anima has also established partnerships with Eli Lilly, Abbvie, and Takeda while remaining independent. “We are well-funded in comparison to some [techbio] companies,” Slonim notes, explaining that the company built its lab through pharma collaborations rather than traditional venture capital. “We never published or disclosed how much money we got in the press, but there are more things that happen as part of extending them or milestones.”
When asked about the future of drug discovery, Slonim returns to his central theme: “AI needs data. AI is all about the data. So where do I generate data? The right answer is from the disease itself. Go and experiment; do very large-scale experimentation, and experimentally generate new data that nobody has seen before.”
Filed Under: Data science, Drug Discovery, machine learning and AI
