←
Daniel Marbach46 posts
Daniel Marbach avatar

Daniel Marbach

@danmarbach
◷ Joined March 2015▧ 9 Photos and videos
Daniel Marbach avatar
Daniel Marbach@danmarbach·30 Aug 2019

Proud of our @DR_E_A_M Challenge paper dissecting methods to identify disease modules in molecular networks nature.com/articles/s41592-0… Joint work of >400 participants, data providers and organizers... collaborative science at its best, honored to be part!

Daniel Marbach avatar
Daniel Marbach@danmarbach·30 Aug 2019

nature.com/articles/s41592-0… Lesson #4: First, identify modules in each network individually, without merging them. Methods attempting to reveal integrated modules across networks failed to significantly improve predictions, likely because our networks were not sufficiently related.

Daniel Marbach avatar
Daniel Marbach@danmarbach·30 Aug 2019

nature.com/articles/s41592-0… Lesson #3: Whenever possible, diverse networks should be leveraged, as they comprise complementary types of module. Signaling, protein-protein interaction and gene co-expression networks revealed the most disease modules.

Daniel Marbach avatar
Daniel Marbach@danmarbach·30 Aug 2019

nature.com/articles/s41592-0… Lesson #2: Use resulting modules from different methods as is, without forming consensus/ensemble predictions. In Challenge, consensus outperformed individual teams #WisdomOfCrowds. But only when integrating >20 methods, which is not practical for users.