As announced in my last post, here is part two of the BSDB-BSCB Spring Meeting Report. It deals with two presentations on networks of transcription factors (TFs). During development, such dynamic networks of TFs and signaling molecules establish and maintain the spatio-temporal patterns of gene expression characteristic for the developing tissue. Using high throughput approaches in this post-genomics era, we now have the opportunity to comprehensively analyze and model these networks and try to link network features to function.
Part 2: Transcription Factor Networks
Eileen Furlong (EMBL, Heidelberg, Germany) presented her lab’s approach to globally decipher the combinatorial action of TFs in cis-regulatory modules (CRM). Using ChIP-on-chip with antibodies against five TFs required for mesoderm development in Drosophila, they generated a high-resolution, genome-wide dataset, describing TF occupancy during 10 hours of early embryonic development. They then used a subset of this data along with the corresponding in vivo activity data of characterized enhancers to train a machine learning algorithm. Using this approach, they were able to correctly predict the spatio-temporal expression of CRMs not included in the training set, based solely on their combinatorial TF binding profiles, in 80% of the tested cases.
Marion Walhout (University of Massachusetts Medical School, Worcester, USA) presented her team’s global analysis of 34 basic helix-loop-helix (bHLH) TFs in C. elegans. They set out to measure all the parameters describing a bHLH TF’s function: which bHLH partner it dimerizes with; where and when the TF is expressed; which DNA sequences it predominantly binds to; and whether it might preferentially regulate genes involved in certain processes. They then combined all of these parameters for every bHLH TF in an integrated network, from which they predicted and experimentally confirmed the function of a specific bHLH, and systematically compared the parameters among all possible bHLH pairs. These analyses linked certain TF dimers to specific processes and interlinked a subset of dimers with each other, uncovering overlapping and specific functions.
I found these presentations inspiring as they combined both previously and newly generated data to try to move beyond merely looking for patterns towards attempting to predict the behavior of the system. However, since these data sets and models are highly complex, it is not always possible to uncover clear-cut trends or rules of behavior.
In part three, my final post on this meeting, I will cover talks on several topics: Stem cells, limb development and evo-devo.