Cellular State Transformations Using Deep Learning for Precision Medicine Applications.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Identifying gene expression differences between normal and diseased tissues is crucial for understanding disease mechanisms and developing targeted therapies.
• Current methods may struggle with identifying subtle or patient-specific expression alterations, especially from limited sample sizes.

Purpose of the Study:

• To introduce a novel deep learning method, the Transcriptome State Perturbation Generator (TSPG), for identifying genomic expression changes between distinct tissue states.
• To demonstrate TSPG's capability in detecting biologically relevant expression patterns in human cancer tissues.
• To showcase TSPG's utility in precision medicine for identifying patient-specific transcriptional aberrations.

Main Methods:

• Developed a deep learning model utilizing generative adversarial networks (GANs).
• TSPG learns the necessary transcriptome perturbations to transition from a source (e.g., tumor) to a target (e.g., normal) gene expression profile.
• Applied TSPG to RNA-sequencing data from human kidney cancer biopsy samples.

Main Results:

• TSPG effectively identified biologically relevant alternate expression patterns between normal and tumor human tissue samples.
• Demonstrated the ability of TSPG to detect patient-specific differentially expressed genes when comparing an individual's tumor sample to a healthy kidney gene expression profile.
• Successfully applied TSPG to a single-patient biopsy sample, highlighting its potential in scenarios with limited replication.

Conclusions:

• TSPG is a novel and effective deep learning approach for identifying transcriptome state perturbations.
• The method facilitates the discovery of patient-specific gene expression aberrations in cancer.
• TSPG offers a promising technique for precision medicine, aiding in the identification of potential targeted therapies.