Related Experiment Video
Updated: Oct 17, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
7.0K
Artificial image objects for classification of breast cancer biomarkers with transcriptome sequencing data and
Xiangning Chen1, Daniel G Chen2, Zhongming Zhao3,4
1410 AI, LLC, Germantown, MD, 20876, USA. va.samchen@gmail.com.
Breast Cancer Research : BCR
|October 11, 2021
Summary
This study introduces a novel method to convert RNA sequencing data into artificial image objects (AIOs), enabling accurate classification of cancer biomarkers like Ki67 and Nottingham histologic grade (NHG) using convolutional neural networks (CNNs). This approach effectively handles complex genomic data for improved cancer patient care.
Area of Science:
- Bioinformatics
- Computational Biology
- Genomics
Background:
- Transcriptome sequencing data is widely available but challenging to use in clinical settings due to high dimensionality and gene correlations.
- Effective utilization of RNA sequencing data for clinical applications remains a significant hurdle.
Purpose of the Study:
- To develop a novel method for transforming high-dimensional RNA sequencing data into artificial image objects (AIOs).
- To apply convolutional neural network (CNN) algorithms for classifying cancer biomarkers using these AIOs.
- To improve the clinical utility of transcriptome data for cancer diagnosis and prognosis.
Main Methods:
- RNA sequencing data was transformed into artificial image objects (AIOs), where genes represent pixels and expression levels represent pixel intensity.
- Convolutional neural network (CNN) models were designed and trained using multiple datasets (GSE96058, GSE81538, GSE163882) to classify Ki67 status and Nottingham histologic grade (NHG).
- Fivefold cross-validation and independent testing were performed to evaluate model performance.
Main Results:
- Classification accuracy and AUC for Ki67 status reached 0.821 ± 0.023 and 0.891 ± 0.021, respectively, with cross-validation.
- For NHG, weighted average accuracy was 0.820 ± 0.012 and AUC was 0.931 ± 0.006.
- Models trained on one dataset and tested on another achieved high accuracy for both Ki67 and NHG, outperforming original study results and demonstrating better survival prediction power for Ki67.
Conclusions:
- RNA sequencing data can be effectively transformed into AIOs for classification using CNNs, addressing challenges of high dimensionality and gene correlations.
- The AIO method offers a data-driven, consistent, and automation-ready approach for biomarker classification from RNA sequencing data.
- This technique has the potential to enhance cancer patient care through more efficient and accurate biomarker analysis.
