Related Experiment Video
Updated: Jun 14, 2025
09:53
Quantifying the Brain Metastatic Tumor Micro-Environment using an Organ-On-A Chip 3D Model, Machine Learning, and Confocal Tomography
Published on: August 16, 2020
7.2K
An interpretable survival model for diffuse large B-cell lymphoma patients using a biologically informed visible
Jie Tan1,2, Jiancong Xie1, Jiarong Huang3
1School of Computer Science and Engineering, Sun Yat-sen University, Guangzhou, China.
Computational and Structural Biotechnology Journal
|August 30, 2024
Summary
VNNSurv, a novel survival model for Diffuse Large B-cell Lymphoma (DLBCL), uses a visible neural network to predict patient outcomes and identify impactful genes. This approach improves prognostic accuracy and aids in understanding DLBCL heterogeneity for precision medicine.
Area of Science:
- Oncology
- Bioinformatics
- Computational Biology
Background:
- Diffuse Large B-cell Lymphoma (DLBCL) is the most common non-Hodgkin lymphoma (NHL), known for its significant heterogeneity.
- Existing prognostic models for DLBCL often rely on transcriptomic data, while genetic variation analysis is more clinically prevalent.
- Current subtyping methods for DLBCL may not fully capture its heterogeneity due to a focus on highly mutated genes.
Purpose of the Study:
- To develop a novel survival model for DLBCL patients utilizing a biologically informed visible neural network (VNN).
- To enhance prognostic accuracy and genetic subtyping of DLBCL by integrating genetic variation data.
- To improve the interpretability of DLBCL prognostic models for identifying key genes and pathways influencing patient outcomes.
Main Methods:
- Development of VNNSurv, a survival model based on a visible neural network (VNN) tailored for DLBCL.
- Validation of VNNSurv performance using cross-validation on the HMRN cohort and an external TCGA cohort.
- Identification of high-impact genes, including those with low alteration frequencies, to develop a genetic-based prognostic index (GPI) and subtype identification method.
Main Results:
- VNNSurv achieved an average C-index of 0.72 on the HMRN cohort (n=928), outperforming baseline methods.
- The model demonstrated strong performance on the external TCGA cohort (n=48) with a C-index of 0.70 using the top 30 impactful genes.
- VNNSurv facilitated the identification of impactful genes and pathways, leading to a genetic-based prognostic index (GPI) and subtypes with superior prognostic consistency.
Conclusions:
- VNNSurv is a valuable and interpretable survival model for DLBCL, offering improved prognostic stratification and subtype identification.
- The model's interpretability aids in understanding DLBCL heterogeneity and has significant implications for precision medicine.
- The VNNSurv framework is adaptable and can be extended to develop prognostic models for other diseases.
