AI model may improve patient response to cancer therapy


By Michael Woodhead

4 Jul 2024

Dr Danh-Tai Hoang

A new artificial intelligence (AI) precision oncology tool can help to select the most suitable treatment for cancer patients, according to researchers at The Australian National University (ANU), Canberra

In a paper published  in Nature Cancer (link here), researchers at the university’s Biological Data Science Institute describe how DeepPT deep learning was developed to predict a patient’s genome-wide tumour mRNA expression.

According to lead author Dr Danh-Tai Hoang, DeepPT was then combined with a second tool called ENLIGHT, which predicts response to targeted and immune therapies from the inferred expression values

Used together, the two-step approach was found to successfully predict a patient’s response to cancer therapies across multiple types of cancer.

He said DeepPT was trained on over 5,500 patients across 16 prevalent cancer types, including breast, lung, head and neck, cervical and pancreatic cancers. In testing, it successfully predicted transcriptomics in all 16 The Cancer Genome Atlas cohorts tested and generalised well to two independent datasets.

“We saw an improvement in patient response rate from 33.3% without using our model to 46.5% cent with using our model,” said Dr Hoang.

The ENLIGHT–DeepPT tool successfully predicted true responders in five independent patient cohorts involving four different treatments spanning six cancer types, with an overall odds ratio of 2.28 and a 39.5% increased response rate among predicted responders versus the baseline rate.

Dr Hoang said DeepPT builds on previous work by the same ANU researchers to develop a tool to help classify brain tumours. Both AI tools draw on histopathology images, also providing another key benefit for patients.

“This cuts down on delays in processing complex molecular data, which can take weeks,” Dr Hoang explained.

“Any kind of delay obviously poses a real challenge when dealing with patients with high-grade tumours who might require immediate treatment. In contrast, histopathology images are routinely available, cost-effective and timely.”

“We know that selecting a suitable treatment for cancer patients can be integral to patient outcomes,” added Dr Hoang.

According to the paper, the model works by leveraging hematoxylin and eosin-stained tumour slides.

“Notably, its prediction accuracy, obtained without any training on the treatment data, is comparable to that achieved by directly predicting the response from the images, which requires specific training on the treatment evaluation cohorts,” the researchers noted.

Already a member?

Login to keep reading.

Email me a login link