I do not know if this affects any of the CTC or other blood tests in use for PCa, but it is reason to investigate further if you are considering using one in developing a treatment strategy. Another example that the use of highly-touted large-learning models (AI) does not in any way guarantee good results. In all areas where your health and longevity are at stake best to "look before you leap".
* * *
Cancer blood test concepts may be based on flawed science: WSJ, by Mariah Taylor, Becker's Hospital Review, published online 09/03/24.
Four years ago, a study found that cancer has unique microbial signatures that would allow a blood test to diagnose cancer. But in recent months, the study has faced scrutiny for inaccurate data, The Wall Street Journal reported Aug. 30.
The study's lead author, Rob Knight, PhD, a University of California San Diego professor, is widely regarded as a pioneer of big-data microbial analysis, according to the report. His study was published in Nature and used more than 17,000 samples from 10,000 people with cancer. The research has been cited more than 600 times. Dozens of groups have based new work on the data and even the private sector has taken notice, with several companies attempting to create blood screening tests.
But in June, the paper was retracted following criticism from other scientists who brought up issues with its methodology and findings. In a statement through a university spokesperson, Dr. Knight told the Journal the retraction was warranted but that the paper's major conclusions are true.
However, skeptics say that some of the microbes flagged as components of cancer signatures weren't known to exist in humans.
The "near-perfect association between microbes and cancer types reported in the study is, simply put, a fiction," an analysis published October 2023 in the journal mBio stated. The analysis also found the researchers incorrectly deployed a genomic tool to match tumor data to microbial sequencing.
"It wasn't a close call," Steven Salzberg, PhD, a computational biologist at Baltimore-based Johns Hopkins University, told the Journal. "This data is completely wrong."
When Nature retracted the study, it cited the above critiques and noted that the paper's authors agreed with its retraction. But the retraction has created a ripple effect, with many other studies having to correct or retract their work due to the original piece's flawed science.
* * *
The DISCUSSION section (emphasis added) from the mBio analysis (linked below) is very damning to the original Knight research:
* * *
DISCUSSION
The original findings of a strong association between microbial species and 33 different cancer types were based on a large collection of DNA and RNA sequencing samples taken from human cancers and from matched normal tissues, which in turn was processed by a sophisticated machine-learning method to create highly accurate classifiers that could distinguish among tumor types and could distinguish tumor from normal tissue (4). Many of these classifiers used bacterial and viral genera that were not known to exist in humans, and therefore raised questions about their plausibility (31); however, this observation alone was not a fatal flaw. It did lead us to explore the machine-learning models more closely, though, in an effort to determine why organisms such as non-human extremophile microbes appeared as key features in the classifiers.
After re-analyzing all of the raw and transformed data, and after downloading and re-analyzing the original reads from more than 1,200 tumor and normal samples, we identified two major errors: first, the raw read counts were vastly overestimated for nearly every bacterial species, often by a factor of 1,000 or more. The likely cause of these overestimates was that the metagenomics database included thousands of draft genomes, which are known to be contaminated with human sequences. Consequently, as we showed above, millions of human reads were erroneously assigned to bacterial or archaeal genera. Second, the process of transforming the raw read counts into normalized values erroneously tagged many of the genera with values that were unique to specific cancer types. It is possible that this information leakage occurred during supervised normalization. When these values were fed to machine-learning classifiers, the algorithms discovered these artificial tags and built highly accurate classifiers, often using features (genera) that in the raw data had zero discriminative power. This error seems to have involved every tumor type and many genera that had zero or near-zero read counts across all of the human samples.
Either of these two errors suffices to invalidate the conclusions of the Poore et al. study and of the other studies that relied upon the same data. The original data matrix of raw read counts contained millions of wildly inaccurate values, and the normalized data compounded this error by tagging the cancer types with distinctive normalized values. Our conclusion after re-analysis is that the near-perfect association between microbes and cancer types reported in the study is, simply put, a fiction.
* * *
The ASM Journal's mBio analysis mentioned in the Becker article is here:
Major data analysis errors invalidate cancer microbiome findings, ASM Journals - mBio, Vol. 14, No. 5, Human Microbiome - Research Article, 9 October 2023.
journals.asm.org/doi/10.112...
This is link to Becker article:
Cancer blood test concepts may be based on flawed science: WSJ, by Mariah Taylor, Becker's Hospital Review, published online 09/03/24.
beckershospitalreview.com/o...
And a link to the WSJ article (behind paywall) is here:
The Far-Reaching Ripple Effects of a Discredited Cancer Study - A flawed study attracted grants, investments and other researchers who based new work on the faulty findings, by Nidhi Subbaraman, The Wall Street Journal, Aug. 30, 2024 7:00 am ET.
wsj.com/health/healthcare/c...2
As always, Stay S&W,
Ciao - cujoe
Unmetabolized Estrogen (Depurinated Adducts) How I got PCa and how to prevent and detect it.
Glutamine blockade to starve tumors, Using bacteria in the microbiome to attack cancer, 2 articles on finding dormant cancer cells in bone
The Warburg Effect [an increase in the rate of glucose uptake and preferential production of lactate, even in the presence of oxygen.]