Audrey Woods | MIT CSAIL Alliances
Embodying the visionary spirit core to MIT’s principles, CSAIL Professor Manolis Kellis imagines a dramatically different future for medicine and healthcare. With the emergence of AI technology that allows us to understand highly complex data—such as the immensely multifaceted systems that run the human body—Professor Kellis believes the time is now to collectively rethink how we approach drug development and deployment, moving away from current blunt methods toward precise, targeted, and mechanistically-driven treatments.
Professor Kellis is using AI to understand cellular circuitry at the mechanistic level, connect disease manifestation to biochemical pathways, model drug interactions and effects at microscopic resolution, and ultimately develop precision and personalized treatments addressing causal, precise, mechanistic dysregulations at the root of disease onset. By leveraging these technologies, Professor Kellis envisions a future where personalized therapeutics become a reality by combining pathway-centered treatments tuned to each patient’s specific combination of disease hallmarks. He hopes that by focusing on the molecular underpinnings and causal disease mechanisms, this modular approach will lead to more streamlined drug development, an improved understanding of illness variations and biomarkers of disease progression, and, most importantly, better outcomes for patients.
FINDING MEANING: THE LANGUAGE OF LIFE
Professor Kellis has always wanted to understand the hidden principles underlying the world around us and find the patterns that bring meaning to complexity. Growing up in a family that moved around the world, he was exposed to multiple languages and cultures at a young age, which gave him an appreciation for the diverse ways in which concepts flow across cultures. His interest in the evolution of language led him to study the evolution of genomes—the language of biology—and the flow of genes and evolutionary design principles across time.
As a young scientist, Professor Kellis was “struck by the conundrum of how do you get evolution to move so rapidly when it took a billion years to get single celled organisms and another billion years to get multicellular organisms? How do you, in the blink of an eye in evolutionary terms, get so much complexity to arise?” With a highly quantitative background in mathematics, computer science, and computational geometry, Professor Kellis approached this problem from the stance of an engineer, looking for the driving principles that optimize DNA adaptability over time. “I was fascinated with the concept of ‘evolvability,’ that perhaps what led to this great acceleration is that during the first few billion years, life got much better at evolving, figuring out modularity, hierarchical organization, functional basic building blocks, and the reuse of parts.” He then looked for these evolutionary principles in genetic data directly, seeking to understand the language of evolutionary operations and, through them, the language of DNA and the building blocks of genomes. This includes the genes, the protein domains, the gene-regulatory control signals, the words and grammar of gene regulation, and the parts that are combined and rearranged to yield meaningful output, rather than exhaustive exploration of the solution search space that naïve genetic algorithms would follow.
Since his graduate studies, Professor Kellis has devoted his career to developing computational methods for understanding our genetic makeup and expression. “ Initially, it was comparative genomic signatures to find the building blocks. Then it was epigenomic signatures to see how they light up in different contexts and in different tissues and cell types and conditions. Then it was about understanding genetic mutations associated with disease and how they are perturbing the cellular circuitry.” He wrote some of the first papers showing how genetic mutations were not random in the genome but specifically localized in classes of regulatory elements, offering new understanding for disease pathways and potential treatment.
Now, his lab is building on this foundation to explore a future of personalized medicine. They aim to give providers the ability to understand patients at the genetic level and treat very particular disease expressions using modular, combinable, highly precise treatment options.
A NEW VISION FOR PERSONALIZED THERAPEUTICS
The key challenge tackled by his lab’s current research is that drugs don’t act in isolation, but in the context of the cellular circuitry. The Kellis Lab is addressing this head-on by understanding how each drug affects the cellular networks surrounding its protein targets, how a drug’s effect on each protein percolates through the cellular circuitry, and how these alter the precise cellular signatures between healthy and disease-associated cellular states. For example, his lab showed that Alzheimer’s disease presents very different cellular signatures across individuals of different age, sex, ancestry, biomarkers, and the distinct set of genetic predispositions that each individual carries. In this sense, there are as many “types” of Alzheimer’s as there are patients, making personalized therapeutics untenable and certainly unaffordable.
However, Professor Kellis has shown that despite this dramatic complexity and combinatorial increase in the diversity of disease manifestations, there are only a small number of recurrent signatures, or disease hallmarks. Targeting those hallmarks “flattens out the combinatorics” according to Professor Kellis, enabling researchers and drug developers to develop a drug for each pathway separately and then use the combinatorics to their advantage by providing modular therapeutic combinations, thus matching the modularity of nature itself and the modularity of disease that his methods have uncovered. “By understanding the unique combination of disease hallmarks that each patient has, we can guide a unique therapeutic approach for each patient informed by our mechanistic understanding, prescribe drugs that directly address the root causes of disease for each patient, and recommend combinations that can address multiple dysregulations that a given patient might be afflicted by.”
To guide the next generation of therapeutic development, Kellis and his team are using a four-pronged approach to mechanistic, precision, and personalized therapeutics:
- First, they are building AI foundation models to understand the building blocks of biology and chemistry, with joint modeling of protein structure and function, chemical structure and function, and the impact of drug-protein interactions across biochemical pathways.
- Second, they are building models to understand how individual drug-protein interactions fit in the larger circuitry of human cells, spanning the multiple levels of cellular signatures, disease hallmarks, and complex phenotypes for normal and disease states.
- Third, they combine this knowledge into a “glass box AI” data science workbench that allows visual exploration, manipulation, integration, and automation, where scientists and AI agents collaborate for iterative therapeutic refinement and virtual drug screening.
- Fourth, they model trajectories of disease state across many patients to guide personalized interventions, aggregating genetic, molecular, biomarker, and imaging information into an integrated risk score for each pathway and matching interventions.
“We have already made great strides for each of these components,” Professor Kellis says, “putting us in a unique position to construct complete models of human health and disease, for the first time spanning all hierarchies of biology that influence our health”.
HUMAN-AI COLLABORATION
“The language of biology has traditionally been outside the reach of human understanding”, explains Professor Kellis. “The multiple tiers, the hundreds of players involved, the sheer complexity of interactions, and the foreign nature of molecular processes, guided by the language of genomes, chemistry, and biology. We have no intuition for the scale of the reality that we have evolved to naturally interact with.”
The unique opportunity of human-AI collaboration is what will make this next generation of healthcare possible. By using AI as an innovation partner, where the human and AI are thinking side by side, we can overcome this fundamental limitation of human understanding by translating biological and chemical processes into natural language, interactive visualizations, and human-accessible representations that allow humans to be part of the discovery process.
Professor Kellis only sees this research accelerating as AI technology improves. “AI is no longer simply a predictive analytics tool, but a full-fledged scientific discovery companion. Our foundation models bring deep structural understanding across many scales and many data modalities that would previously be unfathomable to integrate.”
A COALITION SPANNING INDUSTRY AND ACADEMIA
To make his transformative vision a reality, Professor Kellis would like to see a grand coalition of effort, integrating multiple disciplines by bringing together the FDA, big pharma, insurance companies, hospitals, research centers, and healthcare workers to “redefine the way we think about disease.” This requires eliminating the current silos of medical research and instead of thinking of disease monolithically, breaking complexity into individually addressable components. To him, this mechanistic and causal view of medicine will lead to more effective therapeutics, fewer side effects, and fewer misdiagnoses and inaccurate prescriptions to non-responding patients. “Instead of throwing a grenade at the problem, we can send in a commando team and they take out exactly the precise causal factors at hand.”
Neither industry nor academia can make this happen on their own, and each serves a crucial and irreplaceable role in this transformation. Researchers are motivated to learn the mechanistic basis of drugs and treatments, to understand how a disease manifests and the factors at play. Graduate students are extraordinarily creative and uninhibited by traditional ways of thinking, which allows them to explore richer and more diverse solutions. Pharma brings extraordinary knowledge, decades of experience and know-how, and critical patient and trial data. “This coalition between pharma and academia is an extraordinary win-win, and now is the perfect time for it,” Professor Kellis emphasizes. “We have elucidated so many different diseases and biological processes and we have so many pathways and targets that are within reach for treatment. Patient lives can be saved and livelihoods improved, and the potential positive impact on the world for such partnerships is larger than any one of us.”
Learn more about Professor Kellis on his website, CSAIL Page, CSAIL Alliances Lab Tour, or Podcast Episode.