Deep learning framework for Alzheimer's disease severity index (SI) from brain transcriptomic data
The code to reproduce the work reported at
https://www.biorxiv.org/content/10.1101/2021.06.08.447615v2
Deep Learning-Based Brain Transcriptomic Signatures Associated with the Neuropathological and Clinical Severity of Alzheimer’s Disease
Qi Wang, Kewei Chen, Yi Su, Eric M. Reiman, Joel T. Dudley, Benjamin Readhead
doi: https://doi.org/10.1101/2021.06.08.447615
The file structures are as follows:
ROSMAP_trained_model -- deep learning (supervised classification) of the neuropathologically confirmed AD and control subjects (n = 234), and unsupervised dimensionality reduction (UMAP) of the whole cohort subjects (n = 634), based on the RNAseq data from DLPFC tissues of ROSMAP cohort (syn8456629).
DeepType: representative training process for AD + control groups (5 fold cross validation)
UMAP: intermediate layer manifold calculation for the whole cohort, dimensionality reduction of the manifold for a 3D tranjectory, and SI (pseudotime) calculation
IndexGenes: index gene identification
MAYO -- SI calculation based on ROSMAP trained model for two brain regions (TCX and CER, syn8466812). Obtaining manifold representation by forward pass of the trained network, carrying out the UMAP transformation of the existing embedding model from ROSMAP to get the trajectory and SI.
MSBB -- SI calculation based on ROSMAP trained model for four brain regions (FP(Brodmann area 10), IFG(Brodmann area 44), STG(Brodmann area 22), PHG(Brodmann area 36), syn8484987). Obtaining manifold representation by forward pass of the trained network, carrying out the UMAP transformation of the existing embedding model from ROSMAP to get the trajectory and SI.
All the synapse data/results are withheld.