Researchers develop AI-based model to detect carcinogens

Researchers develop AI-based model to detect carcinogens
x

Researchers develop AI-based model to detect carcinogens

Highlights

An Artificial Intelligence model that could recognise the carcinogens (cancer-causing compounds) from the chemical structures. This model utilizes a novel approach that specifically targets the biological and chemical properties associated with known carcinogens.

New Delhi: Researchers from IIIT-Delhi have developed an Artificial Intelligence-based model to detect carcinogens in chemical structures, which could have importance for the pharma industry for the screening of new drugs.

The research on the software, named as Metabokiller, has also been published in Nature Chemical Biology, which is one of the most reputed journals in the field of Chemical Biology, said the researchers.

"Our latest work builds an AI model that could recognise the carcinogens (cancer-causing compounds) from the chemical structures. This model utilizes a novel approach that specifically targets the biological and chemical properties associated with known carcinogens," said Gaurav Ahuja, Assistant Professor, Department of Computational Biology, Indraprastha Institute of Information Technology (IIIT), Delhi. Ahuja, along with Dr Debarka Sengupta, has developed six independent machine learning based models that precisely scan every query compound for carcinogen-associated properties.

Other clinical and research institutes such as IIT-Ropar, Rajiv Gandhi Cancer Hospital and Research Centre, and CSIR-Institute of Genomics and Integrative Biology, New Delhi, have also participated in the research. Ahuja said Metabokiller has vast importance and utilization in the pharma industry for the screening of new drugs.

"Of note, in the last few decades, many FDA-approved drugs have been taken back from the market since they were found to cause cancer. It also has huge importance in the cosmetics and food industries," he added. Ahuja said that the Metabokiller, unlike other software, provides explainability. "Machine learning is a black box approach, where the cause or the reason for the prediction is largely obscure. Metabokiller predicts carcinogens and does provide the underlying, human interpretable reasoning for this prediction," he asserted.

Approximately five per cent of the cancer is heritable while around 95 per cent of the cancer is caused by exposure to carcinogens (compounds) in the environment, he said. Talking about the way forward, Ahuja said, "We are in advanced communication with multiple pharma companies to test our software in a real-world scenario. Moreover, the research team at IIIT-D is presently working on establishing a direct link between carcinogens and the mutations which they cause in the DNA."

Show Full Article
Print Article
Next Story
More Stories
ADVERTISEMENT