AI and Multiomics Identify Cancer Biomarker and Drug Target
AI algorithm uses omics to spot cancer biomarkers for personalized medicine.
Posted December 3, 2022 | Reviewed by Vanessa Lancaster
- Researchers used an AI-enabled multiomics platform to identify novel biomarkers for new therapeutic targets from gene expression signatures.
- The AI algorithm enables drug discovery by identifying therapeutic targets for diseases by analyzing data and relevant genes.
- Identification of novel molecular biomarkers stratifies cancer patients with different survival outcomes to provide tailored treatments.
Artificial intelligence (AI) machine learning has the potential to completely revolutionize not only how diseases are detected but also identify personalized treatment based on a patient’s genomics. A new study published in Cell Death & Disease, a peer-reviewed Nature journal, shows how an AI-enabled multiomics platform can identify novel biomarkers for new therapeutic targets from gene expression signatures in cancer-associated diseases.
“Identification of novel molecular biomarkers stratifying cancer patients with different survival outcomes may provide new opportunities for target discovery and subsequent development of tailored therapies,” wrote the researchers from Insilico Medicine, the University of Copenhagen, and the University of Chicago.
To find these biomarkers, the researchers used Insilico Medicine’s AI-based platform called PandaOmics to find new cancer targets and analyze gene expression mutations in rare DNA repair-deficient disorders. PandaOmics is an AI deep-learning algorithm that reads scientific papers in order to produce a graphical representation of the content. It also enables cross-dataset comparison, data harmonization, and pathway analysis of activation or inhibition.
The AI algorithm enables drug discovery by identifying therapeutic targets for diseases by analyzing data and relevant genes. It draws upon five million omics data samples (transcriptomics, genomics, epigenomics, proteomics, single-cell data), in addition to relevant data from over 3.8 million patents, 30 million published biomedical research, 1.3 million drugs from phase one of the clinical trial to launched phase, 342,000 clinical trials, and three million funded life sciences research grants.
In science, omics refers to the fields of study that end with -omics, such as genomics (study of an organism’s genomes), neurogenomics (study of genetic impact on the nervous system), psychogenomics (applied genomics, and proteomics to understand the impact on the normal and diseased brain and behavior), microbiomics (study of genomes of microorganisms), and connectomics (study of the brain’s neural connections—the connectome).
Other examples of omics include pangenomics (study of all of the genes in a species), lipdomics (study of lipid pathways and network), immunoproteomics (study of proteins and the immune response), glycomics (study of sugars and carbohydrates), pharmacogenomics (study of the genome and drug response), toxicogenomics (study of gene and protein activity in cells or tissues when exposed to toxins), metabolomics (study of the chemical processes of metabolites), transcriptomics (the study of an organisms RNA transcripts–transcriptome), proteomics (study of proteins), epigenomics (study of the whole set of epigenetic modifications on the genetic material of a cell–epigenome), and more.
Multiomics is the integration of a variety of omics into a single analysis. The integration of artificial intelligence machine learning with multiomics data analysis has enabled scientists to rapidly discover novel biomarkers.
“In order to select diseases for subsequent gene expression analysis and identification of the novel cancer biomarkers, we performed hierarchical clustering based on the analysis of common clinical phenotypes that are prevalent in multiple DNA repair diseases,” the researchers wrote. “Notably, we found three major disease clusters covering diverse phenotypes.”
The scientists identified three major disease clusters and selected these rare inherited diseases for further analysis: Louis–Bar syndrome (Ataxia-Telangiectasia), Nijmegen Breakage syndrome, and Werner syndrome. All three are autosomal recessive diseases which mean that two copies of the abnormal gene must be present for traits or the syndrome to develop, according to the U.S. National Institute of Health (NIH).
Louis-Bar syndrome is a rare inherited neurodegenerative disorder that causes severe disability and affects body systems such as the nervous system and immune system. People with Louis-Bar syndrome are at elevated risk for developing immune system cancers and some types of blood cancers, according to St. Jude Children’s Research Hospital.
Nijmegen Breakage syndrome is a rare genetic disorder that presents at birth with unusually small head size (microcephaly), dysmorphic facial features, and short stature. The name comes from the multitude of DNA breakage found in those with the disease. According to the NIH, people with Nijmegen Breakage syndrome have an increased risk of cancer, specifically, developing non-Hodgkin lymphoma, a cancer of the cells of the immune system, and other cancers associated with this syndrome, include brain cancers such as glioma and medulloblastoma, as well as rhabdomyosarcoma–a cancer of the tissues of muscles.
Werner syndrome is a rare disease characterized by premature aging and increased risk of cancers such as skin and thyroid cancers. The most common causes of death for those with Werner syndrome are cancer and heart attacks, per the NIH.
The researchers used gene expression datasets from DNA repair-deficient disorders with an elevated risk of cancer in order to identify biomarkers for frequently dysregulated genes that could be associated with cancer progression. In efforts to spot the cancer-related pathways in the diseases, the team analyzed changes in gene expression profiles, specifically those that showed genes that were dysregulated.
“Notably, CEP135 was the most downregulated gene with a similar pattern of expression across the three DNA repair diseases, suggesting that it may be associated with the shared cancer phenotype,” the scientists reported.
The scientists hypothesized that the CEP135 gene might serve as a predictive biomarker that could classify patients into subgroups with different survival outcomes. To test this, the team performed survival analysis for 33 cancer types from The Cancer Genome Atlas (TCGA) dataset, one of the largest and most comprehensive genomics datasets available with cancer samples from over 11,000 patients spanning a twelve-year period according to the U.S. National Cancer Institute.
This resulted in the discovery that the CEP135 gene may serve as a predictive biomarker for patients with sarcoma–a rare type of cancerous tumor that develops in connective tissue. Honing their analysis on sarcoma patients with high expression of the CEP135 gene and low survival outcomes, the scientists discovered, with the help of the AI algorithm, a list of the 20 most promising genes to target. This list of candidate genes was further narrowed to five genes after experimental verification. From these remaining five, the polo-like kinase 1 (PLK1) gene was the only one that showed a significant decrease in cell growth, making it a potential target for future cancer treatment.
“While further target validation is required, this study demonstrated the potential of in silico-based studies for a rapid biomarker discovery and target characterization,” the scientists concluded.
Copyright © 2022 Cami Rosso All rights reserved.