The Genome Institute of Singapore (GIS), an organization under the Agency for Science, Technology and Research (A*Star), has developed an artificial intelligence-based method known as Variant Network (VarNet) to facilitate the advancement of personalized cancer treatments. By employing deep learning techniques, VarNet can detect cancer mutations within DNA fragments present in tumor samples. Dr Anders Skanderup, the group leader of GIS’ Laboratory of Computational Cancer Genomics, emphasized the importance of identifying these mutations accurately to tailor the most effective treatment for individual patients. This precision medicine approach considers genetic variations and environmental factors when determining suitable drugs for cancer treatment.
Ensuring a high level of accuracy in identifying cancer mutations is crucial. VarNet acts as a mutation caller by analyzing raw DNA sequencing data. Through exposure to millions of real cancer mutations and examples of false ones, VarNet utilizes artificial intelligence (AI) to decipher between real and false mutations. By doing so, VarNet surpasses existing mutation identification algorithms, as published in the peer-reviewed scientific journal Nature Communications in July 2022. Unlike other AI-based methods, VarNet relies on deep learning, enabling it to teach itself the rules of distinguishing mutations with minimal human intervention.
Kiran Krishnamachari, the first author of the research paper and an A*Star Computing and Information Science scholar affiliated with GIS, explained that VarNet can learn to identify mutations from raw data, much like a human expert manually inspecting potential mutations. This methodology, which does not require excessive manual labeling, instills confidence in the system’s ability to learn relevant mutational features when trained on vast sequencing datasets. This approach is time-efficient, as VarNet can perform the task across the entire three billion nucleotides of the human genome in a fraction of the time it would take a human expert.
To train VarNet, over 300 matched normal and tumor genomes from seven cancer types were used, including lung, sarcoma, colorectal, lymphoma, thyroid, liver, and gastric cancers. The training data comprised whole-genome sequenced tumor data from hospitals and research institutes in Singapore, such as the National University Hospital Singapore and the National Cancer Centre Singapore, as well as the United States cancer genomics program – The Cancer Genome Atlas. The source code for VarNet is publicly available, allowing the international research community to utilize and share their findings. Dr Skanderup’s team is also collaborating with others to test the technology in clinical research projects.
Although AI methods like VarNet will not replace human doctors, they can offer doctors more accurate and detailed information to support their decision-making. The goal is to move VarNet into clinics to effectively tailor treatment strategies for patients. In Singapore, cancer is the leading cause of death, accounting for 23.9% of deaths in 2022. The Singapore Cancer Registry Annual Report 2020 revealed that between 2016 and 2020, there were 80,753 reported cases of cancer in the country. The development and implementation of AI technology, such as VarNet, have the potential to improve cancer treatment outcomes in the future.