Who doesn’t want the latest techniques to check your pathology results and tell you, and your doctor what’s wrong? Recent research by Alliance members, the Queensland Genomics Health Alliance together with the CSIRO and partners have funded a project to look at this very problem.
Researchers are working with pathologists to help them more easily separate ‘pathogenic or disease causing’ from ‘non-pathogenic or non-disease causing’ genotypes amongst the millions of genomic locations flagged in a patient’s genomic sequencing record. Stage one of the project is already underway and will be tested in a number of pathology laboratories in Queensland and Victoria. Re-creating the widely used Variant Effect Predictor (VEP) using cloud-native architecture (serverless), will reduce the current bottlenecks in annotating patient genomic data in pathology reports and increase the level of knowledge about disease association and gene function. From a healthcare system perspective, this frees up resources to do more clever things, such as AI. And from a clinical perspective, this enables the pathologist to provide a faster and more precise diagnosis.
Stage two of the project is to use AI methods such as machine learning to help Pathologists make sense of large volumes of information. Specifically, CSIRO’s machine learning library VariantSpark can help assess a genotype using polygenic risk scores, which take resilience or exacerbation factors from across the genome into acccount. Furthermore, as each pathology lab has a proven workflow, another exciting angle of stage two is to use machine learning for learning from historic annotations and predicting new samples automatically in the same style. This creates a continously learning pathology system, which enables doctors to make more informed decisions in real-time through an efficient, explainable and more standardised pathology system.
Individual genomic information tells a lot about your likely health risk factors, but these are not necessarily reflected or involved in your particular disease states. Instead, other influencers including lifestyle, social, economic and environmental factors may be more relevant and need to be taken into account for diagnostics and treatment choices. And, while these tools won’t be putting pathologists or doctors out of business, they will improve the speed and accuracy of a patient’s diagnosis and provide a more personalised healthcare approach.
It’s anticipated the cloud-based system will be available by mid 2020 and will be shipped as a self-installing package for individual pathology labs to run in their own cloud account. This caters for the stringent data privacy and ownership regulations which are likely to come into place in the future.
For more information about this work, contact firstname.lastname@example.org