AI Tool Offers Hope for Faster, More Personalized Acute Myeloid Leukemia Treatment
Acute myeloid leukemia (AML), a rare and aggressive cancer affecting both blood and bone marrow, presents a significant challenge for oncologists. The disease has a high recurrence rate and lacks a standardized treatment approach. Currently, determining the optimal treatment strategy requires analyzing a patient’s unique genetic makeup, a process that can take weeks—time that is critical for AML patients, as the median age of survival after initial diagnosis is less than five years.
However, a new artificial intelligence (AI) tool developed by Kiran Vanaja, assistant research professor in bioengineering at Northeastern University’s Roux Institute, promises to dramatically shorten this diagnostic timeline and pave the way for more personalized treatment plans.
Mapping Genetic Mutations with AI
The AI platform aims to not only diagnose AML but similarly map the diverse genetic mutations driving the disease in each patient. Once these mutations are identified, a computational model—a neural network—suggests potential drugs and predicts the likelihood of drug resistance. This could reduce the time from diagnosis to treatment from weeks to as little as a single night, according to Vanaja.
The ‘LEGO Blocks’ of Life and Deep Learning
Understanding AML requires delving into the intricacies of cellular processes. Vanaja uses the analogy of genes as “LEGO blocks,” explaining that a deep learning neural network can rapidly assemble and analyze countless combinations of these genetic components.
Traditional methods like gene and RNA sequencing provide information about a cell’s internal components, but Vanaja points out that what’s inside a cell doesn’t always reflect its behavior. Cancer cells, originating as specialized stem cells, lose their defined function and exhibit unpredictable adaptations when exposed to therapies. These adaptations create a disconnect between a cell’s genotype (its genetic makeup) and its phenotype (its observable characteristics).
Untangling Genotype and Phenotype
The challenge lies in untangling these mismatches. Considering the approximately 50,000 known genes, even a subset of those present in the human genome creates a vast number of possible combinations. A neural network, inspired by the structure of the human brain, is uniquely equipped to sift through these complexities at the necessary speed.
The neural network operates through interconnected processing layers, similar to neurons, where each layer processes a component of the task and directs the next step. This allows the network to analyze individual genes and calculate the permutations of their expression within cancer cells.
Training the AI Model
Vanaja and his team trained their AI model by feeding it genotype and phenotype data from thousands of cells from approximately a dozen AML patients. They further refined the model using data from scientific studies on AML cells, enhancing its ability to disentangle the complex relationships between genes and their expression in cancer.
Even as initially focused on AML, the tool’s core function—connecting genotype to phenotype—has broader applications. Vanaja believes it could be extended to other cancers and diseases. The next steps involve continued model training with more patient samples and validation against real-world clinical measurements. Future research will also explore its potential for solid tumors.
About Kiran Vanaja
Kiran Vanaja is an assistant research professor in the Life Sciences and Medical Research Group at the Roux Institute at Northeastern University. He also holds a research assistant professor position with the Bioengineering department of Northeastern University in Boston. His research focuses on building computational mathematical models of signal transduction networks implicated in diseases like cancer and type-2 diabetes. Prior to Northeastern, Vanaja completed postdoctoral training at the Systems Biology Institute of Yale University and holds a PhD in biomedical engineering from Johns Hopkins University.