Related Experiment Video
Updated: Jul 23, 2025
08:16
Comparative Lesions Analysis Through a Targeted Sequencing Approach
Published on: November 5, 2019
6.8K
ForestSubtype: a cancer subtype identifying approach based on high-dimensional genomic data and a parallel random
Junwei Luo1, Yading Feng1, Xuyang Wu1
1School of Software, Henan Polytechnic University, Jiaozuo, China.
BMC Bioinformatics
|July 19, 2023
Summary
This study introduces ForestSubtype, a novel method combining random forests and autoencoders to identify new cancer subtypes using gene expression data. This approach enhances personalized cancer treatment by improving subtype classification accuracy.
Area of Science:
- Bioinformatics
- Computational Biology
- Genomics
Background:
- Accurate cancer subtype classification is crucial for personalized cancer treatment.
- Existing methods for classifying cancer subtypes from high-dimensional gene expression data often yield unsatisfactory results.
- Leveraging prior knowledge of well-established cancer subtypes is significant for identifying novel, meaningful subtypes.
Purpose of the Study:
- To develop an effective computational approach for identifying novel cancer subtypes using high-dimensional gene expression data.
- To integrate prior knowledge of cancer subtypes into the classification process.
- To improve the accuracy and reliability of cancer subtype identification.
Main Methods:
- A combined parallel random forest and autoencoder approach, named ForestSubtype, was developed.
- Prior knowledge of cancer subtypes was used to train an initial module and extract candidate features.
- Parallel random forests were employed to compute feature weights, followed by feature selection.
- An autoencoder was utilized to reduce the dimensionality of selected features to two dimensions.
- K-means++ clustering was applied for novel cancer subtype identification.
Main Results:
- ForestSubtype demonstrated superior performance in cancer subtype identification compared to other methods.
- Validation was performed using breast cancer gene expression data from The Cancer Genome Atlas (TCGA) and the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC).
- The generalizability of ForestSubtype was further confirmed using two additional cancer datasets.
Conclusions:
- The integration of high-dimensional gene expression data with parallel random forests and autoencoders, guided by prior knowledge, offers a more effective strategy for identifying new cancer subtypes.
- ForestSubtype represents an advancement in computational methods for cancer research and personalized medicine.
