AITL: Adversarial Inductive Transfer Learning with input and output space adaptation for pharmacogenomics

Area of Science:

• * Computational biology
• * Bioinformatics
• * Precision medicine

Background:

• * Pharmacogenomics aims to predict patient drug response using omics data.
• * Limited clinical data necessitates transfer learning from pre-clinical datasets.
• * Discrepancies in gene expression (input) and drug response measures (output) challenge model generalization.

Purpose of the Study:

• * To develop a novel method for transfer learning in pharmacogenomics.
• * To address both input and output space discrepancies between pre-clinical and clinical data.
• * To improve the accuracy of predicting drug response for precision oncology.

Main Methods:

• * Proposed Adversarial Inductive Transfer Learning (AITL), a deep neural network.
• * Employed adversarial domain adaptation and multi-task learning.
• * Utilized gene expression data from patients and cell lines to predict drug response.

Main Results:

• * AITL effectively addresses input and output discrepancies between datasets.
• * Experimental results show AITL outperforms existing pharmacogenomics and transfer learning methods.
• * The method demonstrates potential for more accurate guidance in precision oncology.

Conclusions:

• * AITL is the first adversarial inductive transfer learning method to tackle both input and output discrepancies.
• * The developed method offers a promising approach for leveraging pre-clinical data in clinical settings.
• * AITL has the potential to significantly advance precision oncology by improving drug response prediction.