Liver Cancer
Second Leading Cancer
Increasing incidence
Increasing mortality
Aggravated by various factors
Heterogeneity is high
Makes prediction challenging
Has 2 to 6 subtypes
Identified by various data types
mRNA
miRNA
DNA methylation
Most studies are for identification
Lacked survival prediction
Proposal to use deep learning framworks
Autoencoders for multi-omics integration
The Cancer Genome Atlas (TCGA)
Dataset used
mRNA expression
miRNA expression
CPG methylation
Clinical information
Methods and Dataasets
TCGA Dataset
Used for label extraction
Used for training a Support Vector Machine (SVM)
Additional validations using multiple cohorts
Chinese cohort
E-TABM-36
NCI cohort
Hawaiian cohort
LIRI-JP Cohort
RNA-Seq
gene expression
miRNA expression
gene expression
DNA methylation array
Used three pre-processed datasets
Feature extraction using autoencoders
Feature Selection
Univariate Cox proportional hazards
log-rank < 0.05
K-means clustering
Metrics and other frameworks
Evaluation metrics
Log-rank P value of Cox-PH regression
Brier Score
Concordance index
Correct prediction model metric
Score around 0.70 = good model
Score around 0.5 = random background
For survival analysis
Measures accuracy of prediction
Other methods
Principal Component Analysis instead of autoencoders
Use of fewer features
Functional Analysis
Clinical covariate analysis
Differential expression
TP53 mutation analysis
Enriched pathways analysis
Frequency distribution analysis
Associations of subtypes
Note: Functional analysis employed may be used for future works
Race
Grade
Gender
Stage
Fisher Exact Tests
Note: Comparison across different cohorts
Results
Based on C-Index the model is robust
Low Brier score indicates low error rates
Significant log rank P scores
Achieved great results from different cohorts
Chaudhary, K., Poirion, O. B., Lu, L., and
Garmire, L. X. Deep learning–based multi-omics
integration robustly predicts survival in liver cancer.
Clinical Cancer Research 24, 6 (2018), 1248–1259.