DeePathology

DeePathology is a deep learning-based tool that utilizes multi-task and transfer learning to simultaneously infer various properties of biological samples from their whole transcriptome data. The key features and advantages of DeePathology are:

1. It encodes the entire transcription profile into a compact 8-dimensional latent vector, capturing the sample's essential information.

2. From this low-dimensional latent space, DeePathology can recover mRNA and miRNA expression profiles and predict the tissue and disease type of the sample.

3. The latent space representation is more effective than the original gene expression profiles for discriminating samples based on their tissue and disease characteristics.

4. DeePathology was trained on a large dataset of 10,750 clinical samples from 34 classes (1 healthy and 33 cancer types) across 27 tissues.

6. For tissues with multiple cancer subtypes, DeePathology achieves a high accuracy of 99.4% in identifying the correct subtype.

7. The tool is robust against noise and missing values in the input data.