Study findings could lead to new clinical laboratory diagnostics that give pathologists a more detailed understanding about certain types of cancer
New studies proving artificial intelligence (AI) can be used effectively in clinical laboratory diagnostics and personalized healthcare continue to emerge. Scientists in the UK recently trained an AI model using machine learning and deep learning to enable earlier, more accurate detection of 13 different types of cancer.
DNA stores genetic information in sequences of four nucleotide bases: A (adenine), T (thymine), G (guanine) and C (cytosine). These bases can be modified through DNA methylation. There are millions of DNA methylation markers in every single cell, and they change in the early stages of cancer development.
One common characteristic of many cancers is an epigenetic phenomenon called aberrant DNA methylation. Modifications in DNA can influence gene expression and are observable in cancer cells. A methylation profile can differentiate tumor types and subtypes and changes in the process often come before malignancy appears. This renders methylation very useful in catching cancers while in the early stages.
However, deciphering slight changes in methylation patterns can be extremely difficult. According to the scientists, “identifying the specific DNA methylation signatures indicative of different cancer types is akin to searching for a needle in a haystack.”
Nevertheless, the researchers believe identifying these changes could become a useful biomarker for early detection of cancers, which is why they built their AI models.
“Computational methods such as this model, through better training on more varied data and rigorous testing in the clinic, will eventually provide AI models that can help doctors with early detection and screening of cancers,” said Shamith Samarajiwa, PhD (above), Senior Lecturer and Group Leader, Computational Biology and Genomic Data Science, Imperial College London, in a news release. “This will provide better patient outcomes.” With additional research, clinical laboratories and pathologists may soon have new cancer diagnostics based on these AI models. (Photo copyright: University of Cambridge.)
The researchers then used a combination of machine learning and deep learning techniques to train an AI algorithm to examine DNA methylation patterns of the collected data. The algorithm identified and differentiated specific cancer types, including breast, liver, lung and prostate, from non-cancerous tissue with a 98.2% accuracy rate. The team evaluated their AI model by comparing the results to independent research.
In their Biology Methods and Protocols paper, the authors noted that their model does require further training and testing and stressed that “the important aspect of this study was the use of an explainable and interpretable core AI model.” They also claim their model could help medical professionals understand “the underlying mechanisms that contribute to the development of cancer.”
Using AI to Lower Cancer Rates Worldwide
According to the Centers for Disease Control and Prevention (CDC), cancer ranks as the second leading cause of death in the United States with 608,371 deaths reported in 2022. The leading cause of death in the US is heart disease with 702,880 deaths reported in the same year.
Globally cancer diagnoses and death rates are even more alarming. World Health Organization (WHO) data shows an estimated 20 million new cancer cases worldwide in 2022, with 9.7 million persons perishing from various cancers that year.
The UK researchers are hopeful their new AI model will help lower those numbers. They state in their paper that “most cancers are treatable and curable if detected early enough.”
More research and studies are needed to confirm the results of this study, but it appears to be a very promising line of exploration and development of using AI to detect, identify, and diagnose cancer earlier. This type of probing could provide pathologists with improved tools for determining the presence of cancer and lead to better patient outcomes.
Findings could lead to new biomarkers clinical laboratories would use for identifying cancer in patients and monitoring treatments
As DNA “dark matter” (the DNA sequences between genes) continues to be studied, researchers are learning that so-called “junk DNA” (non-functional DNA) may influence multiple health conditions and diseases including cancer. This will be of interest to pathologists and clinical laboratories engaged in cancer diagnosis and may lead to new non-invasive liquid biopsy methods for identifying cancer in blood draws.
This technique could enable non-invasive monitoring of cancer treatment and cancer diagnosis, Technology Networks noted.
“Our study shows that ARTEMIS can reveal genomewide repeat landscapes that reflect dramatic underlying changes in human cancers,” said study co-leader Akshaya Annapragada (above), an MD/PhD student at the Johns Hopkins University School of Medicine, in a news release. “By illuminating the so-called ‘dark genome,’ the work offers unique insights into the cancer genome and provides a proof-of-concept for the utility of genomewide repeat landscapes as tissue and blood-based biomarkers for cancer detection, characterization, and monitoring.” Clinical laboratories may soon have new biomarkers for the detection of cancer. (Photo copyright: Johns Hopkins University.)
Detecting Early Lung, Liver Cancer
Artemis is a Greek word meaning “hunting goddess.” For the Johns Hopkins researchers, ARTEMIS also describes a technique “to analyze junk DNA found in tumors” and which float in the bloodstream, Financial Times explained.
“It’s like a grand unveiling of what’s behind the curtain,” said geneticist Victor Velculescu, MD, PhD, Professor of Oncology and co-director of the Cancer Genetics and Epigenetics Program at Johns Hopkins Kimmel Cancer Center, in the news release.
“Until ARTEMIS, this dark matter of the genome was essentially ignored, but now we’re seeing that these repeats are not occurring randomly,” he added. “They end up being clustered around genes that are altered in cancer in a variety of different ways, providing the first glimpse that these sequences may be key to tumor development.”
ARTEMIS could “lead to new therapies, new diagnostics, and new screening approaches for cancer,” Velculescu noted.
Repeats of DNA Sequences Tough to Study
For some time technical limitations have hindered analysis of repetitive genomic sequences by scientists.
“Genetic changes in repetitive sequences are a hallmark of cancer and other diseases, but characterizing these has been challenging using standard sequencing approaches,” the study authors wrote in their Science Translational Medicine paper.
“We developed a de novok-mer (short sequences of DNA)-finding approach called ARTEMIS to identify repeat elements from whole-genome sequencing,” the researchers wrote.
The scientists put ARTEMIS to the test in laboratory experiments.
The first analysis involved 1,280 types of repeating genetic elements “in both normal and tumor tissues from 525 cancer patients” who participated in the Pan-Cancer Analysis of Whole Genomes (PCAWG), according to Technology Networks, which noted these findings:
A median of 807 altered elements were found in each tumor.
About two-thirds (820) had not “previously been found altered in human cancer.”
Second, the researchers explored “genomewide repeat element changes that were predictive of cancer,” by using machine learning to give each sample an ARTEMIS score, according to the Johns Hopkins news release.
The scoring detected “525 PCAWG participants’ tumors from the healthy tissues with a high performance” overall Area Under the Curve (AUC) score of 0.96 (perfect score being 1.0) “across all cancer types analyzed,” the Johns Hopkins’ release states.
Liquid Biopsy Deployed
The scientists then used liquid biopsies to determine ARTEMIS’ ability to noninvasively diagnose cancer. Researchers used blood samples from:
ARTEMIS classified patients with lung cancer with an AUC of 0.82.
ARTEMIS detected people with liver cancer, as compared to others with cirrhosis or viral hepatitis, with a score of AUC 0.87.
Finally, the scientists used their “ARTEMIS blood test” to find the origin of tumors in patients with cancer. They reported their technique was 78% accurate in discovering tumor tissue sources among 12 tumor types.
“These analyses reveal widespread changes in repeat landscapes of human cancers and provide an approach for their detection and characterization that could benefit early detection and disease monitoring of patients with cancer,” the researchers wrote in Science Translational Medicine.
Large Clinical Trials Planned
Velculescu said more research is planned, including larger clinical trials.
“While still at an early stage, this research demonstrates how some cancers could be diagnosed earlier by detecting tumor-specific changes in cells collected from blood samples,” Hattie Brooks, PhD, Research Information Manager, Cancer Research UK (CRUK), told Financial Times.
Should ARTEMIS prove to be a viable, non-invasive blood test for cancer, it could provide pathologists and clinical laboratories with new biomarkers and the opportunity to work with oncologists to promptly diagnosis cancer and monitor patients’ response to treatment.
Scientists believe useful new clinical laboratory assays could be developed by better understanding the huge number of ‘poorly researched’ genes and the proteins they build
Researchers have added a new “-ome” to the long list of -omes. The new -ome is the “unknome.” This is significant for clinical laboratory managers because it is part of an investigative effort to better understand the substantial number of genes, and the proteins they build, that have been understudied and of which little is known about their full function.
The Unknome Database includes “thousands of understudied proteins encoded by genes in the human genome, whose existence is known but whose functions are mostly not,” according to a news release.
The database, which is available to the public and which can be customized by the user, “ranks proteins based on how little is known about them,” the PLOS Biology paper notes.
It should be of interest to pathologists and clinical laboratory scientists. The fruit of this research may identify additional biomarkers useful in diagnosis and for guiding decisions on how to treat patients.
“These uncharacterized genes have not deserved their neglect,” said Sean Munro, PhD (above), MRC Laboratory of Molecular Biology in Cambridge, England, in a press release. “Our database provides a powerful, versatile and efficient platform to identify and select important genes of unknown function for analysis, thereby accelerating the closure of the gap in biological knowledge that the unknome represents.” Clinical laboratory scientists may find the Unknome Database intriguing and useful. (Photo copyright: Royal Society.)
Risk of Ignoring Understudied Proteins
Proteomics (the study of proteins) is a rapidly advancing area of clinical laboratory testing. As genetic scientists learn more about proteins and their functions, diagnostics companies use that information to develop new assays. But did you know that researchers tend to focus on only a small fraction of the total number of protein-coding DNA sequences contained in the human genome?
The study of proteomics is primarily interested in the part of the genome that “contains instructions for building proteins … [which] are essential for development, growth, and reproduction across the entire body,” according to Scientific American. These are all protein-coding genes.
Proteomics estimates that there are more than two million proteins in the human body, which are coded for 20,000 to 25,000 genes, according to All the Science.
To build their database, the MRC researchers ranked the “unknome” proteins by how little is known about their functions in cellular processes. When they tested the database, they found some of these less-researched proteins important to biological functions such as development and stress resistance.
“The role of thousands of human proteins remains unclear and yet research tends to focus on those that are already well understood,” said Sean Munro, PhD, MRC Laboratory of Molecular Biology in Cambridge, England, in the news release. “To help address this we created an Unknome database that ranks proteins based on how little is known about them, and then performed functional screens on a selection of these mystery proteins to demonstrate how ignorance can drive biological discovery.”
In the paper, they acknowledged the human genome encodes about 20,000 proteins, and that the application of transcriptomics and proteomics has “confirmed that most of these new proteins are expressed, and the function of many of them has been identified.
“However,” the authors added, “despite over 20 years of extensive effort, there are also many others that still have no known function.”
They also recognized limited resources for research and that a preference for “relative safety” and “well-established fields” are likely holding back discoveries.
The researchers note “significant” risks to continually ignoring unexplored proteins, which may have roles in cell processes, serve as targets for therapies, and be associated with diseases as well as being “eminently druggable,” Genetic Engineering News reported.
Setting up the Unknome Database
To develop the Unknome Database, the researchers first turned to what has already come to fruition. They gave each protein in the human genome a “knownness” score based on review of existing information about “function, conservation across species, subcellular localization, and other factors,” Interesting Engineering reported.
It turns out, 3,000 groups of proteins (805 with a human protein) scored zero, “showing there’s still much to learn within the human genome,” Science News stated, adding that the Unknome Database catalogues more than 13,000 protein groups and nearly two million proteins.
The researchers then tested the database by using it to determine what could be learned about 260 “mystery” genes in humans that are also present in Drosophila (small fruit flies).
“We used the Unknome Database to select 260 genes that appeared both highly conserved and particularly poorly understood, and then applied functional assays in whole animals that would be impractical at genome-wide scale,” the researchers wrote in PLOS Biology.
“We initially selected all genes that had a knownness score of ≤1.0 and are conserved in both humans and flies, as well as being present in at least 80% of available metazoan genome sequences. … After testing for viability, the nonessential genes were then screened with a panel of quantitative assays designed to reveal potential roles in a wide range of biological functions,” they added.
“Our screen in whole organisms reveals that, despite several decades of extensive genetic screens in Drosophila, there are many genes with essential roles that have eluded characterization,” the researchers conclude.
Clinical Laboratory Testing Using the Unknome Database
Future use of the Unknome Database may involve CRISPR technology to explore functions of unknown genes, according to the PLOS Biology paper.
Munro told Science News the research team may work with other research efforts aimed at understanding “mysterious proteins,” such as the Understudied Proteins Initiative.
The Unknome Database’s ability to be customized by others means researchers can create their own “knownness” scores as it applies to their studies. Thus, the database could be a resource in studies of treatments or medications to fight diseases, Chemistry World noted.
According to a statement prepared for Healthcare Dive by SomaLogic, a Boulder, Colorado-based protein biomarker company, diagnostic tests that measure proteins can be applied to diseases and conditions such as:
“The 27-protein model has potential as a ‘universal’ surrogate end point for cardiovascular risk,” the researchers wrote in Science Translational Medicine.
Proteomics definitely has its place in clinical laboratory testing. The development of MRC-LMB’s Unknome Database will help researchers’ increase their knowledge about the functions of more proteins which should in turn lead to new diagnostic assays for labs.
Research findings could lead to new biomarkers for genetic tests and give clinical laboratories new capabilities to diagnose different health conditions
New insights continue to emerge about “junk DNA” (aka, non-coding DNA). For pathologists and clinical laboratories, these discoveries may have value and eventually lead to new biomarkers for genetic testing.
One recent example comes from researchers at Stanford Medicine in California who recently learned how non-coding DNA—which makes up 98% of the human genome—affects gene expression, the function that leads to observable characteristics in an organism (phenotypes).
The research also could lead to a better understanding of how short tandem repeats (STRs)—the number of times a gene is copied into RNA for protein use—affect gene expression as well, according to Stanford.
“We’ve known for a while that short tandem repeats or STRs, aren’t junk because their presence or absence correlates with changes in gene expression. But we haven’t known how they exert these effects,” said study lead Polly Fordyce, PhD (above), Associate Professor of Bioengineering and Genetics at Stanford University, in a news release. The research could lead to new clinical laboratory biomarkers for genetic testing. (Photo copyright: Stanford University.)
To Bind or Not to Bind
In their Science paper, the Stanford researchers described an opportunity to explore a new angle to transcription factors binding to some sequences, also known as sequence motifs.
“Researchers have spent a lot of time characterizing these transcription factors and figuring out which sequences—called motifs—they like to bind to the most,” said the study lead Polly Fordyce, PhD, Associate Professor of Bioengineering and Genetics at Stanford University, in a Stanford Medicine news release.
“But current models don’t adequately explain where and when transcription factors bind to non-coding DNA to regulate gene expression. Sometimes, no transcription factor is attached to something that looks like a perfect motif. Other times, transcription factors bind to stretches of DNA that aren’t motifs,” the news release explains.
Transcription factors are “like light switches that can turn genes on or off depending on what cells need,” notes a King’s College LondonEDIT Labblog post.
But why do transcription factors target some places in the genome and not others?
“To solve the puzzle of why transcription factors go to some places in the genome and not to others, we needed to look beyond the highly preferred motifs,” Fordyce added. “In this study, we’re showing that the STR sequence around the motif can have a really big effect on transcription factor binding, providing clues as to what these repeated sequences might be doing.”
Such information could aid in understanding certain hereditary conditions and diseases.
“Variations in STR length have been associated with changes in gene expression and implicated in several complex phenotypes such as schizophrenia, cancer, autism, and Crohn’s disease. However, the mechanism by which STRs affect transcription remains unknown,” the researchers wrote in Science.
Special Assays Explore Binding
According to their paper, the research team turned to the Fordyce Lab’s previously developed microfluidic binding assays (MITOMI, k–MITOMI, and STAMMP) to analyze the impact of different DNA sequences on transcription factor binding.
“In the experiment we asked, ‘How do these changes impact the strength of transcription factor binding?’ We saw a surprisingly large effect. Varying the STR sequence around a motif can have a 70-fold impact on the binding,” Fordyce wrote.
In an accompanying Science article titled, “Repetitive DNA Regulates Gene Expression,” Thomas Kuhlman, PhD, Assistant Professor, Physics and Astronomy, University of California, Riverside, wrote that the study “demonstrates that STRs exert their effects by directly binding transcription factor proteins, thus explaining how STRs might influence gene expression in both normal and diseased states.”
“This research unveils, for the first time, the intricate connection between how variants in the non-coding genome affect genes that are associated with blood pressure and with hypertension. What we’ve created is a kind of functional map of the regulators of blood pressure genes, “said Philipp Maass, PhD, Lead Researcher and Assistant Professor Molecular Genetics, University of Toronto, in a news release.
The research team used massively parallel reporter assay (MPRA) technology to analyze 4,608 genetic variants associated with blood pressure.
The findings could aid precision medicine for cardiovascular health and may possibly be adopted to other conditions, according to The Hospital for Sick Children.
“The variants we have characterized in the non-coding genome could be used as genomic markers for hypertension, laying the groundwork for future genetic research and potential therapeutic targets for cardiovascular disease,” Maass noted.
Why All the ‘Junk’ DNA?
Clinical laboratory scientists may wonder why genetic research has primarily focused on 20,000 genes within the genome, leaving the “junk” DNA for later investigation. So did researchers at Harvard University.
“After the Human Genome Project, scientists found that there were around 20,000 genes within the genome, a number that some researchers had already predicted. Remarkably, these genes comprise only about 1-2% of the three billion base pairs of DNA. This means that anywhere from 98-99% of our entire genome must be doing something other than coding for proteins,” they wrote in a blog post.
“Imagine being given multiple volumes of encyclopedias that contained a coherent sentence in English every 100 pages, where the rest of the space contained a smattering of uninterpretable random letters and characters. You would probably start to wonder why all those random letters and characters were there in the first place, which is the exact problem that has plagued scientists for decades,” they added.
Not only is junk DNA an interesting study subject, but ongoing research may also produce useful new biomarkers for genetic diagnostics and other clinical laboratory testing. Thus, medical lab professionals may want to keep an eye on new developments involving non-coding DNA.
Technology could enable patients to monitor their own oxygen levels and transmit that data to healthcare providers, including clinical laboratories
Clinical laboratories may soon have a new data point to add to their laboratory information system (LIS) for doctors to review. Researchers have determined that smartphones can read blood-oxygen levels as accurately as purpose-built pulse oximeters.
Conducted by researchers at the University of Washington (UW) and University of California San Diego (UC San Diego), the proof-of-concept study found that an unmodified smartphone camera and flash along with an app is “capable of detecting blood oxygen saturation levels down to 70%. This is the lowest value that pulse oximeters should be able to measure, as recommended by the US Food and Drug Administration,” according to Digital Health News.
This could mean that patients at risk of hypoxemia, or who are suffering a respiratory illness such as COVID-19, could eventually add accurate blood-oxygen saturation (SpO2) readings to their lab test results at any time and from any location.
“In an ideal world, this information could be seamlessly transmitted to a doctor’s office. This would be really beneficial for telemedicine appointments or for triage nurses to be able to quickly determine whether patients need to go to the emergency department or if they can continue to rest at home and make an appointment with their primary care provider later,” Matthew Thompson, DPhil, Professor of Global Health and Family Medicine at University of Washington, told Digital Health News. Clinical laboratories may soon have a new data point for their laboratory information systems. (Photo copyright. University of Washington.)
UW/UC San Diego Study Details
The researchers studied three men and three women, ages 20-34. All were Caucasian except for one African American, Digital Health News reported. To conduct the study, a standard pulse oximeter was placed on a finger and, on the same hand, another of the participant’s fingers was placed over a smartphone camera.
“We performed the first clinical development validation on a smartphone camera-based SpO2 sensing system using a varied fraction of inspired oxygen (FiO2) protocol, creating a clinically relevant validation dataset for solely smartphone-based contact PPG [photoplethysmography] methods on a wider range of SpO2 values (70–100%) than prior studies (85–100%). We built a deep learning model using this data to demonstrate an overall MAE [Mean Absolute Error] = 5.00% SpO2 while identifying positive cases of low SpO2 < 90% with 81% sensitivity and 79% specificity,” the researchers wrote in NPJ Digital Medicine.
When the smartphone camera’s flash passes light through the finger, “a deep-learning algorithm deciphers the blood oxygen levels.” Participants were also breathing in “a controlled mixture of oxygen and nitrogen to slowly reduce oxygen levels,” Digital Health News reported.
“The camera is recording a video: Every time your heart beats, fresh blood flows through the part illuminated by the flash,” Edward Wang, PhD, Assistant Professor of Electrical and Computer Engineering at UC San Diego and senior author of the project, told Digital Health News. Wang started this project as a UW doctoral student studying electrical and computer engineering and now directs the UC San Diego DigiHealth Lab.
“The camera records how much that blood absorbs the light from the flash in each of the three color channels it measures: red, green, and blue. Then we can feed those intensity measurements into our deep-learning model,” he added.
The deep learning algorithm “pulled out the blood oxygen levels. The remainder of the data was used to validate the method and then test it to see how well it performed on new subjects,” Digital Health News reported.
“Smartphone light can get scattered by all these other components in your finger, which means there’s a lot of noise in the data that we’re looking at,” Varun Viswanath, co-lead author in the study, told Digital Health News. Viswanath is a UW alumnus who is now a doctoral student being advised by Wang at UC San Diego.
“Deep learning is a really helpful technique here because it can see these really complex and nuanced features and helps you find patterns that you wouldn’t otherwise be able to see,” he added.
Each round of testing took approximately 15 minutes. In total the researchers gathered more than 10,000 blood oxygen readings. Levels ranged from 61% to 100%.
“The smartphone correctly predicted whether the subject had low blood oxygen levels 80% of the time,” Digital Health News reported.
Smartphones Accurately Collecting Data
The UW/UC San Diego study is the first to show such precise results using a smartphone.
“Other smartphone apps that do this were developed by asking people to hold their breath. But people get very uncomfortable and have to breathe after a minute or so, and that’s before their blood-oxygen levels have gone down far enough to represent the full range of clinically relevant data,” said Jason Hoffman, a PhD student researcher at UW’s UbiComp Lab and co-lead author of the study.
The ability to track a full 15 minutes of data is a prime example of improvement. “Our data shows that smartphones could work well right in the critical threshold range,” Hoffman added.
“Smartphone-based SpO2 monitors, especially those that rely only on built-in hardware with no modifications, present an opportunity to detect and monitor respiratory conditions in contexts where pulse oximeters are less available,” the researchers wrote.
“This way you could have multiple measurements with your own device at either no cost or low cost,” Matthew Thompson, DPhil, Professor of Global Health and Family Medicine at University of Washington, told Digital Health News. Thompson is a professor of both family medicine and global health and an adjunct professor of pediatrics at the UW School of Medicine.
What Comes Next
The UW/UC San Diego research team plans to continue its research and gather more diversity among subjects.
“It’s so important to do a study like this,” Wang said. “Traditional medical devices go through rigorous testing. But computer science research is still just starting to dig its teeth into using machine learning for biomedical device development and we’re all still learning. By forcing ourselves to be rigorous, we’re forcing ourselves to learn how to do things right.”
Though no current clinical laboratory application is pending, smartphone use to capture biometrics for testing is increasing. Soon, labs may need a way to input all that data into their laboratory information systems. It’s something to consider.
