New Delhi: What causes cancer? Researchers at Delhi’s Indraprastha Institute of Information Technology may have moved a step towards answering that all important question through an AI platform that helps decode DNA damage and what may have led to the malignancy.
‘MutAIverse’ is the first of its kind generative AI platform anywhere in the world, the researchers said. It was tested on 27 people suffering from head and neck cancer in Guwahati, Assam, and went to the root to trace the exposure that possibly contributed to the disease — chewing smokeless tobacco.
Though AI adds a layer of information in understanding cancer development, it does not in its current form conclusively prove that DNA damage due to the exposure caused the cancer, cautioned experts. The researchers include those from Guwahati’s National Institute of Pharmaceutical Education and Research (NIPER) and Council of Scientific and Industrial Research (CSIR)-Institute of Genomics and Integrative Biology (IGIB).
Their research focuses attention on the causative factors rather than just the diagnosis and can be a useful screening tool, the study authors said.
“Nobody talks about what caused the cancer. What were you exposed to? Were you exposed to pollution? Were you exposed to contaminated water? Or were you exposed to cigarettes? Nobody tells you,” lead author Gaurav Ahuja, associate professor at the department of computational biology, IIIT, told PTI.
The environment we live in continuously exposes us to multiple chemicals, which can induce mutations in the DNA, he explained.
“In my point of view, the study is a step forward when people are talking not about the fact that one got cancer, but in a way that they can identify what they were exposed to, so that family members can be protected, if the exposure is prevalent in the locality they live in,” Ahuja said.
The genAI platform also helped expand existing reference libraries of DNA adducts — toxic chemicals that can attach to and cause changes to one’s DNA, potentially causing cancer — from fewer than 400 to over three lakh. The AI system is described in a paper published in the Journal of Cheminformatics.
“The platform is the first-of-its-kind in the world where generative AI has helped open up the largest library of potential reference DNA adducts (for cancer),” Ahuja said.
MutAIverse, a contraction of ‘mutational universe’, features a machine-learning-based backtracking tool called AdductLinker.
Upon finding a damaged DNA in a cancer patient’s tumour, AdductLinker works backward like a “detective tracing footprints”, allowing us to pinpoint the exact environmental chemical — genotoxin or carcinogen — that originally caused the cancer, Ahuja explained.
The team experimentally validated MutAIverse using biopsy samples of head and neck cancer patients with a history of consuming smokeless tobacco.
The samples were collected in collaboration with the Dr. Bhubaneswar Borooah Cancer Institute in Guwahati. DNA adductomics — a comprehensive study of DNA adducts — was performed at NIPER.
Generative AI, a type of artificial intelligence trained on massive amounts of data, can create original content, including images, audios and videos, and respond to a user’s request in the natural language (a prompt).
For MutAIverse, a prompt is the mass spectrometry data from the patient’s tumour biopsy and output is characteristics of the patient’s exposure. Mass spectrometry is a diagnostic method that breaks down cell samples into molecular profiles, which carry information about DNA mutations.
“The DNA is made up of four nucleotide bases or building blocks — the letters A, T, G and C. Now, a chemical entering your body can perform mutations, which means changing the letters of the DNA and composition of A, T, G and C. So, A is no more A, it will be something else,” Ahuja explained.
“MutAIverse basically allows you a platform, where you can isolate the patient’s DNA samples, read the A-T-G-C and even if there is a 0.1 per cent abnormality, it back traces and identifies the exposure the patient had,” he said.
Output of the generative AI is a comprehensive breakdown of the patient’s exposure, pointing out the exact environmental chemical or toxin the patient was exposed to — along with a confidence score — and showing the specific structure of the damaged DNA. The output also contains a list of intermediate chemicals formed in the body due to a breakdown of the toxin.
“Application to in-house generated DNA adductomics data from tissue biopsies of smokeless tobacco-induced head and neck cancer patients revealed (a) selective enrichment of both novel and known DNA adducts,” the authors wrote.
MutAIverse also uses simulations to predict how toxic chemicals break down inside human cells and physically mutate the DNA.
“If a particular chemical, for example, a tobacco product, or occupational exposure is identified, the patient can try to avoid further exposure, and people with similar risks can also be warned,” said Saravanan Matheshwaran, associate professor at the Indian Institute of Kanpur’s (IIT-K) department of biological sciences and bioengineering, told PTI.
There’s a lot of work that remains to be done.
Discussing the findings, Matheshwaran said, “Finding a DNA adduct only tells us that a chemical has interacted with DNA. It does not prove that this particular damage caused a mutation or eventually caused cancer. Some types of DNA damage may be repaired by the cell, while others may remain and contribute to (a) genomic instability.”
“The AdductLinker approach is particularly useful because it attempts to connect a DNA adduct with the chemical exposure that may have produced it,” he said.
However, detecting an adduct does not by itself prove that it caused the cancer.
“More experimental, clinical and epidemiological evidence would be needed to establish such a connection,” Matheshwaran said.
In India, head and neck cancers are driven overwhelmingly by smokeless tobacco and betel nut chewing, particularly in the north east where the study collected samples from, added Kedar Natarajan, a computational biologist and professor of biology at Ashoka University in Sonepat.
“Knowledge of causative cancer exposure is critical towards understanding cancer severity, mutational landscape, cancer management and potentially (the) treatment regime,” he said.
He added that while the genAI platform makes for an impressive computational tool for basic research, in its current form, it is not a clinical diagnostic tool for hospital or community deployment.
The platform also does not prove that the particular exposure or DNA damage caused a mutation, eventually leading to cancer, the experts said.
Overall, the study provides a complementary layer of information and the platform after validation could be used along with genome sequencing and other approaches to build a better picture of how a particular cancer developed, the experts said.
