Improving Drug Response Prediction in Glioblastoma Using Transfer Learning
Author Information
Author(s): Jie Ju, Ntafoulis Ioannis, Klein Michelle, JT Reinders Marcel, Lamfers Martine, Stubbs Andrew P, Yunlei Li
Primary Institution: Erasmus MC Cancer Institute, University Medical Center Rotterdam
Hypothesis
Can a two-step transfer learning framework enhance drug response prediction accuracy in glioblastoma with small datasets?
Conclusion
The two-step transfer learning framework significantly improved the prediction accuracy of drug response in glioblastoma cell cultures compared to traditional methods.
Supporting Evidence
- The two-step transfer learning framework outperformed traditional deep learning models without transfer learning.
- Using oxaliplatin as a source drug improved prediction accuracy for temozolomide response in glioblastoma.
- The study demonstrated that transfer learning can effectively mitigate the small sample size problem in drug response prediction.
Takeaway
Researchers found a way to better predict how glioblastoma cells respond to a cancer drug by using knowledge from other cancer data, even when they had very few samples to work with.
Methodology
A two-step transfer learning framework was constructed, where deep learning models were pretrained on a large source dataset and then refined on a smaller target dataset.
Potential Biases
Potential biases may arise from the selection of datasets and the inherent variability in drug response among different cancer types.
Limitations
The study is limited by the small sample size of the target dataset and the variability in drug response measurements across different datasets.
Participant Demographics
The study focused on treatment-naïve patients with glioblastoma.
Statistical Information
P-Value
0.010
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website