Predicting Response Prognosis in Pediatric Cancers

Division Of Clinical Sciences - Nci

Investigators

Linked publications & trials

Abstract

Neuroblastomas are cancers that originate from neural crest cells, and their prognosis varies based on factors like age at diagnosis, stage, histology, MYCN amplification, chromosomal ploidy, and the 1p36 deletion status. However, the specific molecular mechanisms determining good or poor prognosis in this and other malignancies remain largely unknown. Our groundbreaking research has shown that gene expression profiling using cDNA microarrays and sophisticated pattern recognition algorithms, such as Artificial Neural Networks, can aid in diagnosing cancers. To further advance this knowledge, the Oncogenomics Section has collaborated with the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) group. Together, we are conducting extensive genomic analysis, employing next-generation whole genome, exome, and transcriptome sequencing on clinically annotated neuroblastoma samples. Through these cutting-edge techniques, we aim to identify somatic mutations and tumor-specific expression patterns that can uniquely pinpoint patients with poor prognoses, as well as those linked to specific genetic aberrations, including MYCN amplification. Rhabdomyosarcoma (RMS) is the most prevalent soft tissue sarcoma in children. Unfortunately, despite aggressive therapy, patients with metastatic or recurrent disease face a poor 5-year survival rate, and there are currently no genomic markers available for risk stratification. To address this critical gap, we launched an international consortium study. The primary objective is to determine the prevalence of driver mutations and their association with clinical outcomes in RMS patients. Tumor samples collected from participants in Children's Oncology Group (COG) and United Kingdom trials (Malignant Mesenchymal Tumour and RMS2005) were subjected to custom capture sequencing. By identifying mutations, indels, gene deletions, and amplifications, we plan to perform survival analysis and create a searchable companion database (https://clinomics.ccr.cancer.gov/clinomics/public/). This database will include comprehensive genomic variants and clinical annotations, including survival data. The goal is to uncover genomic profiles that correlate with prognosis and to identify the specific genes responsible for these biological characteristics. Following our report that specific genomic alterations in RMS, strongly correlated with survival, we utilized convolutional neural networks (CNN) to learn histologic features associated with driver mutations and outcome using hematoxylin and eosin (H&E) images of RMS. Digital whole slide H&E images were collected from clinically annotated diagnostic tumor samples from 321 patients with RMS enrolled in Children's Oncology Group (COG) trials (1998-2017). Patches were extracted and fed into deep learning CNNs to learn features associated with mutations and relative event-free survival risk. The performance of the trained models was evaluated against independent test sample data (n = 136) or holdout test data. The trained CNN could accurately classify alveolar RMS, a high-risk subtype associated with PAX3/7-FOXO1 fusion genes, with an ROC of 0.85 on an independent test dataset. CNN models trained on mutationally-annotated samples identified tumors with RAS pathway with a ROC of 0.67, and high-risk mutations in MYOD1 or TP53 with a ROC of 0.97 and 0.63, respectively. Remarkably, CNN models were superior in predicting event-free and overall survival compared with current molecular-clinical risk stratification. This study demonstrates that high-risk features, including those associated with certain mutations, can be readily identified at diagnosis using deep learning. CNNs are a powerful tool for diagnostic and prognostic prediction of rhabdomyosarcoma, which will be tested in prospective COG clinical trials. In parallel, we are utilizing advanced proteomic analysis techniques like mass spectroscopy and phospho-proteomic analysis. These methods allow us to quantitatively measure protein expression levels and phosphorylation status in different cell types and tissues. By comparing proteins from samples with poor (death) and good (event-free survival 3 years) outcomes, we plan to identify 3000-4000 differentially expressed proteins. This data will help us identify potential targets for therapy, diagnostic markers, and prognostic indicators for high-risk patients. Additionally, it will provide valuable insights into the biology of these tumors, which currently show resistance to conventional therapy. In summary, our research endeavors are aimed at revolutionizing cancer diagnostics, prognostics, and treatment by unraveling the molecular complexities that influence prognosis in neuroblastomas and rhabdomyosarcomas. Through innovative genomic, proteomic analyses, and machine learning, we hope to unlock valuable information that will ultimately lead to more effective and personalized approaches for patients facing these challenging diseases.