Photo Credit: ArtemisDiana
EMBER, a novel embedding approach, integrates diverse transcriptomic datasets for enhanced precision in breast cancer diagnostics and treatment.
Transcriptomics has significantly advanced biomedical research, particularly in breast cancer subtyping and diagnostics, according to research presented online in NPJ Breast Cancer. Despite its potential, the clinical application of transcriptomic data remains limited due to various challenges, often due to the need for single-sample predictors.
To address these care gaps, molecular EMBeddER (EMBER), a novel embedding approach that integrates a unified space of 11,000 breast cancer transcriptomes, allows for phenotype prediction on a single-sample basis. This methodology accurately identifies the five molecular subtypes of breast cancer by leveraging key biological pathways, such as estrogen receptor signaling, cell proliferation, DNA repair, and epithelial-mesenchymal transition.
Understanding What’s Different
The authors explained that researchers used four independent patient cohorts, including the POETIC trial, to validate EMBER’s robustness and clinical relevance. This demonstrated EMBER’s capacity to capture clinical responses to endocrine therapy and to identify mechanisms underlying therapy resistance, such as increased androgen receptor signaling and decreased TGFβ signaling. Notably, EMBER’s estrogen receptor (ER) signaling score outperforms the traditional immunohistochemistry-based ER index, offering a superior predictive tool for selecting patients for endocrine therapy.
Breast cancer is the most commonly diagnosed cancer globally, with over 70% of cases being estrogen receptor-positive (ER+). According to ASCO guidelines, these patients typically receive endocrine therapy. High-throughput transcriptomic profiling has refined breast cancer subtyping, leading to clinically approved RNA-based molecular signatures like Breast Cancer Index, MammaPrint, EndoPredict, Oncotype DX, and Prosigna. These signatures aid in identifying patients who are ER+ that would benefit from chemotherapy, thereby preventing overtreatment. However, they offer limited insight into the biological reasons that necessitate additional therapy and do not evaluate the benefits of endocrine therapy.
With the decreasing costs of global transcriptomics, platforms like the SCAN-B consortium and the Hartwig Medical Foundation can now provide mutational and gene expression-based biomarkers within a week of tumor surgery. These platforms have shown that RNA sequencing can determine ER status and intrinsic molecular subtypes, identify targetable mutations, and calculate risk scores using single-sample predictor algorithms. This suggests that RNA sequencing may replace more expensive predictive testing while offering comprehensive information.
Interpreting Information and Improving Care
EMBER integrates microarray and RNA sequencing datasets into a common space, accommodating individual patient samples. This approach offers a new way to interpret molecular subtypes as a continuum and to identify candidate biological pathways of resistance to endocrine therapy. Unlike traditional batch-effect removal algorithms or embedding techniques, EMBER allows for adding single samples and calculating single sample scores, providing a high-resolution view of molecular subtypes.
Further, EMBER’s integration methodology has demonstrated stability and robustness, capturing key features of breast cancer, such as ER status and intrinsic molecular subtypes. The approach embraces the continuous nature of molecular characteristics, moving away from the limitations of discrete subtype classifications. EMBER’s mapping relies on known biological pathways, and its findings suggest potential benefits of combination treatments for early-stage breast cancer patients, particularly those involving targeted therapies.
The authors noted EMBER’s ability to integrate diverse transcriptomic datasets into a common, clinically relevant space marks a significant step forward in breast cancer diagnostics and treatment.
“Previous attempts at data integration across distinct technical platforms were restricted solely to cell lines and did not encompass single-sample-based approaches, the study authors concluded.
“Here, we have overcome these problems in the early breast cancer setting with the data integration method EMBER using large public patient datasets such as TCGA, METABRIC, SCAN-B, and other cohorts.”
