Domain alignment method based on masked variational autoencoder for predicting patient anticancer drug response.

Area of Science:

• Computational biology
• Genomics
• Pharmacogenomics

Background:

• Accurate prediction of patient response to anticancer drugs is crucial for personalized medicine.
• Discrepancies between cell line and patient data limit the performance of predictive models.
• Existing transfer learning methods attempt domain alignment but can be improved.

Purpose of the Study:

• To develop a robust method for predicting patient anticancer drug response by addressing data distribution differences.
• To leverage knowledge from cell line data for improved patient-specific predictions.
• To introduce a novel domain alignment approach using masked variational autoencoders.

Main Methods:

• Proposed a masked variational autoencoder domain alignment (MVAEDA) method.
• Employed multiple variational autoencoders (VAEs) and mask predictors for feature extraction.
• Utilized generative adversarial training for learning domain-invariant features from gene expression data.

Main Results:

• MVAEDA demonstrated superior performance in predicting anticancer drug response on clinical and preclinical datasets.
• The method effectively aligned features between cell line and patient data domains.
• Experimental results indicate improved accuracy compared to existing state-of-the-art methods.

Conclusions:

• MVAEDA offers a promising solution for bridging the gap between cell line and patient data in drug response prediction.
• The developed domain alignment strategy enhances the generalizability of predictive models.
• This approach holds potential for advancing personalized cancer therapy through more accurate drug response forecasting.