Network Version Information
This page tracks version history for Mavatar Discovery's DINA (tissue- and disease-specific) networks, including changes to network construction methodology.
Version 2
Topic | Version 1 | Version 2 |
|---|---|---|
Edge filtering | Not applied | Edges included only where confidence based normalized T is higher than the median normalized T across all edges. If the median confidence-based score is 0 then all zero values are discarded. Thereafter, median/zero filtering Benjamini–Hochberg adjusted p ≤ 0.05 filtering is applied. |
K nearest neighbors | Based on the top 20 neighbors (see Edge prioritization below) , while additionally ensuring that each protein coding gene has at least K other protein coding neighbors, or maximum 40 neighbors. | Based on the top 50 neighbors, irrespective of gene type. |
Edge prioritization | Based on raw (non-adjusted) T values, evaluated over the full correlation vector. | Based on a confidence-bound score derived from normalized T values, evaluated over the pool of edges passing the two filtering steps above. |
Edge list orientation impact | The neighbors were prioritized per gene, after which duplicated edges (i.e. where two genes prioritized each other) were reduced to only one edge per gene pair | The K nearest neighbors are prioritized for the genes in alphabetical order, ensuring unique edge identification for each gene.
|
Version 1
Implementation of DINA, tissue- and disease-specific networks. These networks were constructed based on our standard correlation method, as well as our AI – AutoEncoder method. See Data Collection and Pre-ProcessingData Collection and Pre-Processing, DINA Network ConstructionDINA Network Construction, and Mavatar Discovery Tools – Scientific Backgroundx for detailed descriptions of our approach.