Related Experiment Video
Updated: Jun 11, 2026
VDJ-Seq: Deep Sequencing Analysis of Rearranged Immunoglobulin Heavy Chain Gene to Reveal Clonal Evolution Patterns of B Cell Lymphoma
Published on: December 28, 2015
Deep learning based individualized cross-platform molecular subtype classification of B-lineage acute lymphoblastic
Bowen Cui1, Huiying Sun2, Shuang Zhao2
1Key Laboratory of Pediatric Hematology & Oncology Ministry of Health, Department of Hematology & Oncology, Biomedical Data Science Center, Pediatric Translational Medicine Institute, Shanghai Children's Medical Center, School of Medicine, Shanghai Jiao Tong University, Shanghai, China. xcxiongmao@126.com.
We developed BALL6, a deep learning tool for classifying B-cell acute lymphoblastic leukemia (B-ALL) subtypes using gene expression. This framework improves diagnostic accuracy and efficiency for leukemia classification.
Area of Science:
- Computational biology
- Genomics
- Machine learning in medicine
Background:
- Accurate molecular subtyping of B-cell acute lymphoblastic leukemia (B-ALL) is critical for effective clinical treatment.
- Current multi-omic diagnostic approaches face challenges in sensitivity, cost, and turnaround time due to rapidly emerging subtypes.
- There is a need for efficient and robust methods for B-ALL subtyping.
Purpose of the Study:
- To develop a robust deep learning framework, BALL6 (B-cell Acute Lymphoblastic Leukemia Subtype Identification based on gene eXpression), for cross-platform B-ALL subtyping.
- To create integrated models capable of distinguishing between B-ALL, T-ALL, and AML, and identifying specific molecular subtypes within B-ALL.
- To enhance the framework's performance on data-limited or imbalanced datasets and ensure robustness to technical noise and missing values.
Main Methods:
- Utilized a recurrent neural network trained on rank-transformed gene expression values to capture subtype signals and minimize technical noise.
- Implemented a rank-based augmentation framework to improve performance on challenging datasets.
- Developed two integrated models: an Acute Leukemia (AL) model and a B-ALL molecular subtype model.
Main Results:
- Achieved high accuracy on independent datasets: 99.38% for the AL model and 93.84% for the B-ALL model on unseen data.
- Demonstrated robustness to missing values, enabling reliable predictions from sparse gene expression profiles, including single-cell RNA sequencing (scRNA-seq) data.
- BALL6 showed significant potential for cross-platform application in leukemia diagnostics.
Conclusions:
- BALL6 provides a robust, accurate, and efficient deep learning framework for B-ALL subtyping and general acute leukemia classification.
- The open-sourced nature and accessible web tool facilitate broad adoption in leukemia research and clinical diagnostics.
- This approach addresses the limitations of current methods, offering a sensitive, cost-effective, and rapid diagnostic solution.
