A new machine learning tool has been developed by bio-engineers at Georgia Tech. It can analyze a patients RNA and match them with the best chemo-drugs. It has matched patients to the best outcome 80% of the time.
The team at Georgia Tech behind the tool – John McDonald, Evan Clayton and Fredrik Vannberg are shown with sequencing equipment in Georgia Tech’s Petit Institute for Bioengineering and Bioscience
The tool works by matching up a patients ‘RNA expression’ to existing info about patient outcomes with specific drugs. It can then predict which chemotherapy drug has provided the best outcome 80% of the time.
Up to now doctors know what first-line chemotherapy drugs to use to treat cancers but it can be tricky to know what to use next if that treatment fails.
The tool is designed to help in selecting those treatments and the team believe its accuracy can be improved by also factoring in additional patient info (like family histories and demographics).
By looking at RNA expression in tumours, we believe we can predict with high accuracy which patients are likely to respond to a particular drug… This information could be used, along with other factors, to support the decisions clinicians must make regarding chemotherapy treatment John McDonald, a professor in the Georgia Tech School of Biological Sciences and director of its Integrated Cancer Research Center
The team built the software using existing records of RNA from tumours, drugs used and the patient outcome.
The team had 152 records available and split this into two parts. They used 114 records to train their software, and then the remaining 38 records to test it. Initially, they focused on ovarian cancer but were able to expand their data to include other cancer types (like liver, breast and pancreatic cancers) that used the same type of drugs.
The software was then able to produce a chart to show the outcome of each drug on a patients specific type of cancer.
Over the last few years, the cost of RNA sequencing has been decreasing. If the trend continues it will cost less to sequence a patients cancer RNA than a mammogram.
The team are planning to make the tool available as open source software. The team believe that an open source approach offers the best way to get the algorithm into clinical use.
As soon as hospitals and cancer treatment centres start using it its accuracy should improve. The more data it can crunch the more accurate it should become.
To really get this into clinical practice, we think we’ve got to open it up so that other people can try it, modify if they want to, and demonstrate its value in real-world situations… We are trying to create a different paradigm for cancer therapy using the kind of open source strategy used in internet technologyJohn McDonald, a professor in the Georgia Tech School of Biological Sciences and director of its Integrated Cancer Research Center
What’s more open sourcing the tool will mean that others can review it and spot improvements and bugs. The tool will be released as a Github repository later this year.
Published as Machine learning predicts individual cancer patient responses to therapeutic drugs with high accuracy in Scientific Reportsvolume 8, Article number: 16444 (2018)