OncoNetExplainer, based on neural networks (NN) and VGG16 networks with a GradCAM++, take patients' Gene Expressions (GE) from The Cancer Genome Atlas(TCGA) as input to return most significant biomarkers and ranked top genes across different cancer types based on mean absolute impact (MAI) and shows that GE is useful for predicting cancer types with high confidence. OncoNetExplainer first embeds high dimensional RNA-Seq data into 2D images and trains CNN and VGG16 networks with GradCAM++ activated to classify 33 tumor types based on patients’ GE profiles and provides a human-interpretable explanation to identify important biomarkers, which is further validated based on the annotations from TumorPortal. Prediction accuracy achieved is up to 96%
Input variables : Patients' Gene Expressions
Output Variables : Most significant biomarkers, top genes across different cancer types based on mean absolute impact
Statistical | : | Somers D | Accuracy | Precision and Recall | Confusion Matrix | F1 Score | Roc and Auc | Prevalence | Detection Rate | Balanced Accuracy | Cohen's Kappa | Concordance | Gini Coefficent | KS Statistic | Youden's J Index |
Infrastructure | : | Log Bytes | Logging/User/IAMPolicy | Logging/User/VPN | CPU Utilization | Memory Usage | Error Count | Prediction Count | Prediction Latencies | Private Endpoint Prediction Latencies | Private Endpoint Response Count |
Visit Model : arxiv.org
Additional links : arxiv.org
Model Category | : | Public |
Date Published | : | April, 2020 |
Healthcare Domain | : |
Life Sciences
Medical Technology |
Code | : | github.com |
Health Risk Management |
Disease Detection |