Acute myeloid leukemia, or AML, is a rare and aggressive cancer that can affect people of all ages. Kiran Vanaja, an assistant research professor in bioengineering at Northeastern University, says that AML also has a high recurrence rate and no one-size-fits-all treatment option.
Because AML impacts both blood and bone marrow, oncologists need samples of both through blood draws and a bone marrow aspiration to determine the disease’s particular genetic makeup and which treatment might be most appropriate. It can often take a month or more after diagnosis for those with AML to start receiving potentially life-saving treatment. The cancer’s “median age of survival is less than five years after initial diagnosis,” according to Vanaja, so time is of the essence.
Based out of Northeastern’s Roux Institute in Portland, Maine, Vanaja was recently awarded a patent for an artificial intelligence tool that, he hopes, will save time.
The new tool includes a platform that may help oncologists not only diagnose the disease but also map out the many different genetic mutations in a given patient that might have caused their AML. Once the mutations are mapped, the platform also includes a computational model known as a neural network that can both suggest potential drugs for an individual patient and also determine the likelihood of the patient developing resistance to those drugs.
The tool could help cut the preliminary time from diagnosis to treatment from weeks down to a single night, Vanaja says.
The LEGO blocks of life
One of the challenges to understanding what might be happening with AML is understanding what’s happening inside the cancerous cells. And to test their new tool, Vanaja’s team had to go back to the basics of cell biology.
02/26/26 – BOSTON, MA. – Kiran Vanaja
Assistant Research Professor, Bioengineering poses for a portrait on Feb. 26, 2026. Photo by Matthew Modoono/Northeastern University
02/26/26 – BOSTON, MA. – Kiran Vanaja
Assistant Research Professor, Bioengineering poses for a portrait on Feb. 26, 2026. Photo by Matthew Modoono/Northeastern University
Vanaja likens genes to LEGO blocks. A deep learning neural net, like the kind he employs to diagnose AML, can assemble many permutations of those blocks and analyze the results in a short amount of time. Photos by Matthew Modoono/Northeastern University
Using conventional strategies — which are still the gold standard, he notes — such as through gene sequencing, RNA sequencing or other techniques, researchers might gain information about what’s inside the cell. But with cancer, Vanaja says, what’s inside the cell doesn’t always align with how the cell acts.
Cells don’t start out as cancer cells, but as basic stem cells that later specialize into different kinds of tissue, whether of the lungs, liver, arteries or something else.
But when a cell becomes cancerous, Vanaja says, it loses that primary specialization, as the team also learned when they treated a bunch of AML cells with FDA-approved therapies.
Many of the cancer cells died off, but after a couple of weeks, Vanaja and his team looked at the surviving cells and observed that they had changed in radical ways. “The cells undergo massive rewiring when subjected to these cancer therapeutics, because they’re literally trying to survive by turning on anything and everything possible.”
These desperate attempts at survival end up creating a mismatch between the cell’s original internal components, its genotype, and how it expresses itself externally, its phenotype.
But figuring out what these mismatches are and untangling them is a gargantuan task.
If we think of the approximately 50,000 genes we know of across all species like a huge LEGO set, each block fits together and layers on top of one another to express in unique ways, Vanaja says. The ways that the genes stack can change that expression, too. Even if we only consider half those genes — which is approximately how many are in the human genome — the combinatorial options would be huge.
A deep learning neural network, a type of artificial intelligence computational model, he says, with its interconnecting processing layers, is the only way to sift through all those combinations at the speed they hope to achieve.
A neural network is inspired by the way the brain works, with processing nodes, not unlike neurons, layered on top of each other. In a neural network, a node in one layer will process a small component of the larger task before deciding which node in the next layer to send the next step of the task to.
This is what makes it so good at the genetic modeling Vanaja needed: A neural net could look at one gene at a time, if need be, and quickly calculate each permutation of how the gene was being expressed by the cancer cell.
To harness this power for their work, Vanaja and his team built a basic AI model by feeding it genotype and phenotype data from thousands of cells from about a dozen patients with AML. Then, they trained and refined the model on all the scientific studies they could find that reported AML cell data so that the model could more accurately disentangle the many possible combinations of genes and how cancer cells were expressing them.
Vanaja continues that this neural network tool isn’t limited to AML or even to cancer, since the AI network’s primary function is to connect the cell’s genotype to its current phenotypic expression. AML was a good starting point for their research, he says, given the cancer’s complexity.
Once samples are taken from a patient, the tool can run its predictions and make treatment suggestions in the space of a single night, Vanaja notes. The next step is to continue training the model on further samples and compare those results against real-world measurements.
Upcoming research, Vanaja continues, will also explore the tool’s “application to solid tumors.”
Noah Lloyd is the assistant editor for research at Northeastern Global News and NGN Research. Email him at n.lloyd@northeastern.edu. Follow him on X/Twitter at @noahghola.