Notes from the Lab: A Better Way to Decontaminate Samples for Early Cancer Detection
The Korem Lab
We study what human-associated microbial communities (the “microbiome”) tell us about their host and how they affect its health. We aim to develop personalized microbiome-based and -guided therapeutics and diagnostics.
“Contamination source modeling with SCRuB improves cancer phenotype prediction from microbiome data,” published in Nature Biotechnology, work co-led with Liat Shenhav, PhD from Rockefeller University.
The cancer problem we are solving
Using a combination of DNA sequencing and machine learning, a prior study found unique microbial signatures in blood for certain types of cancer. Essentially, the results suggest that blood-borne microbial DNA could possibly be used to classify patients with cancer versus those without, as well as discriminate between cancer types. A microbiome-based oncology diagnostic tool that only requires a patient blood sample has the potential to detect cancer earlier and easier than traditional methods.
However, the tool performed well for certain types of cancer — namely prostate cancer and lung cancer — but not for melanoma. We believe this disparity may have to do with issues of sample contamination.
A bit of background
Microbes are everywhere, including where we don’t want them to be. Say we draw blood from a cancer patient and want to know what microbes are in that sample. Instead, we might be getting microbes from the reagents that we’re using in the experiment, or from the skin of the person who is processing the samples. Another issue is that biological material tends to flow from one sample to another. These phenomena are broadly termed “contamination.”
To combat erroneous test results, we developed a new method called Source-tracking for Contamination Removal in microBiomes (SCRuB), a method for high-precision decontamination of microbial data using control samples. Essentially, SCRuB incorporates shared information across multiple samples and controls to precisely identify and remove contamination.
What this new research uncovers
We compared SCRuB against other decontamination methods using plasma samples from patients with lung cancer, prostate cancer, and melanoma. Data processed by SCRuB displayed the strongest predictive performance for melanoma, significantly higher than alternative methods. We were further able to use SCRuB to develop predictors for immunotherapy treatment response using microbial DNA for the melanoma itself. For other cancer types, SCRuB performed comparably to other methods.
Our results demonstrate the importance of decontamination in revealing biological signals that may be masked by contamination, as well as that SCRuB is superior to alternative methods in certain scenarios.
We are now working on improving the capabilities of SCRuB. For example, while most studies nowadays collect contamination controls, these do not capture every type of contamination. So, we're currently developing a method to actually perform decontamination even without having controls on hand. We are further applying SCRuB for development of diagnostic models in other clinical settings, as well as for studying the effect of microbes in the tumor microenvironment.