Uploaded image for project: 'The Virtual Brain'
  1. The Virtual Brain
  2. TVB-1727

Collaboration with MSc in Life Science Informatics

    Details

    • Type: Task
    • Status: Closed
    • Priority: Critical
    • Resolution: Fixed
    • Affects Version/s: None
    • Fix Version/s: 1.4.1
    • Component/s: None
    • Labels:
      None
    • Epic Link:
    • Sprint:
      BRCO - r1.4.1

      Description

      Good morning,

      let me add a little context, in particular for Lia and Marmaduke. I did not have time to brief them yet, since I was travelling (in fact, am still at the airport in Paris). I also added Jochen into this discussion.

      Some time ago Martin and I started a discussion on the possibility go generating a multi scale brain model that combines the powers of data mining approaches and generative mechanistic models. Here below are some words that have been previously written in this context:
      "Generative models are the only means to establish causality of aetipathogenetic pathways and systematically explore novel therapeutic directions. In the best case, generative models include mechanistic multiscale descriptions of brain network systems that can be personalized for a patient and generate biologically realistic time resolved data mimicking the empirical data sets obtained from invasive and non-invasive human brain imaging. A complete multiscale (spatial and temporal) description of the human brain is still decades away, even though massive European investments (e.g. the FET flagship Human Brain Project) are prioritizing this development. State of the art today are data mining approaches informed by prior knowledge derived from meta analysis, literature mining and semantic analysis procedures, which identify links in heterogeneous data leading to subspaces highly associated with certain phenotypes, but are by construction incapable of predicting modalities outside of the data space.

      We wish to combine the predictive power of generative brain network models with the multiscale parameter structures and phenomenological correlations obtained from data mining taking advantage of the respective strengths of both approaches. Rather than mechanistically modeling every level of organization (from the molecular to the non-invasive brain imaging level), we inform the mechanistic brain network models by the outcomes of sophisticated data mining, which spans the links across heterogeneous data types and thus traverses the scales of organization. This novel hybrid concept will accelerate our current way of thinking in multiscale modeling and develop new technologies to identify patterns of alteration across different levels of biological organisation, suggesting new diagnostic indicators and drug targets, facilitating the selection of subjects for clinical trials, providing the data required for disease modelling and simulation, and facilitating the translation of knowledge about the brain from the laboratory to the clinic. "

      I think this will give you an impression on where the journey shall evolve. To be useful, such a hybrid approach needs not only to connect the various levels (genome, epigenome, proteome, etc), but also establish functional relationships, that means as a change occurs on one level, we can predict how the other levels will be affected. This functional relationship does not have to be perfect and complete, but it needs to exist at least partially. Why? TVB makes the direct link to non-invasive personalized brain imaging data. Any alterations of structural/functional nature we can parametrically map out and validate. Changes in brain imaging data due to pathology will be reflected in parameter changes of the mechanistic model. Here comes the second challenge: the model parameters in the various brain regions need to be interpreted and linked to data and knowledge. There will probably be not a systematic way in the beginning, but parameters and relations will repeat themselves, hence we will establish a catalogue and atlas as time evolves. This will take time, but will be doable.

      In a nutshell: we will attempt to cheat nature and develop a multi-level mechanistic approach that does not necessitate a biophysiological causal derivation starting from the molecular level (à la Human Brain Project) but rather replaces this by our hybrid approach, mechanistic on the level of the large-scale brain network where non-invasive brain imaging connects to the model, and relational data and knowledge approach across all other levels.

      I propose to keep it as simple as possible in the beginning. As Martin pointed out, a mapping of data and knowledge on the brain regions and visualization thereof would be a first step.

      Best wishes,

      Viktor

        Attachments

        1. AD-specific_Mechanism_Story_Documentation.pdf
          467 kB
        2. AnandhiBRCO.owl
          13 kB
        3. B_highlight_on_mouse_over_right_area.png
          B_highlight_on_mouse_over_right_area.png
          618 kB
        4. BRCO_0305.owl
          4.81 MB
        5. BRCO_merged_test.xml
          60 kB
        6. C_show_brco_connected_regions.png
          C_show_brco_connected_regions.png
          781 kB
        7. Screen Shot 2015-03-24 at 19.26.06 (2).png
          Screen Shot 2015-03-24 at 19.26.06 (2).png
          352 kB
        8. TVB error message.jpg
          TVB error message.jpg
          706 kB
        9. TVBOnto.dtd
          0.5 kB
        10. workplan_BRCO.docx
          4.64 MB

          Activity

            People

            • Assignee:
              lia.domide Lia Domide
              Reporter:
              lia.domide Lia Domide
            • Votes:
              0 Vote for this issue
              Watchers:
              1 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved:

                Time Tracking

                Estimated:
                Original Estimate - 1 week, 4 days, 2 hours
                1w 4d 2h
                Remaining:
                Remaining Estimate - 0 minutes
                0m
                Logged:
                Time Spent - 1 week, 3 days, 3 hours Time Not Required
                1w 3d 3h