Write-up of the Predictive Modelling Competition Results

[edwintse]

I am currently in the process of writing up a summary of the results that we obtained from the first modelling competition (#538) for my thesis and for the eventual paper on this work.

To get this started it would be super helpful if the entrants from the competition (@spadavec, @holeung, @gcincilla, @kellerberrin, @IamDavyG, @jon-c-silva) were able to provide a brief summary description about the models you developed for a broader audience to understand.

There is a Google Sheet here where a new column could be added for this description (in addition to the Summary column already there).

[holeung]

Great! I'm out of town at the moment but will get to it when I return.

[gcincilla]

Very good. How long should be the description? If you can give us the max number of words/characters as a reference, maybe the descriptions will be more homogeneous.

[edwintse]

Probably around 30 words for each description. It would be good to have information about each model tabulated for the paper.

[kellerberrin]

Hi Edwin,

Thanks for your email. As requested Please find below a brief description of my entry.

The rankings are based on an ensemble of Deep Neural Network models trained on molecular information obtained from the pharmacophore Dragon freeware..

The Dragon software converted compound molecular SDF records (SMILES) into 1666 relevant pharmacophore fields for all compounds.

These fields were normalized and submitted to a conventional (dense) Deep Neural Network (Keras/TensorFlow) for training and analysis using the example compounds provided.

The DNNs were classifiers and were trained on the 3 reported PfATP4 activity classes; [ACTIVE, PARTIAL, INACTIVE] for each of the example compounds.

The trained DNNs exhibited a number of different optimal classifications (local minina) for the example compounds during training.

So the different optimal classification DNNs (without further training) were then combined into an ensemble using a small Neural Network meta-classifier to produce a final ranking of the test compounds.

This ranking was expressed as the probability that a compound would be in the [ACTIVE] activity class.

I hope this brief description satisfies your request. Please contact me if you have any further queries about the above.

Best Regards,

James.

On Thu, 30 May 2019 at 23:09, Edwin Tse notifications@github.com wrote:

I am currently in the process of writing up a summary of the results that we obtained from the first modelling competition (#538 https://github.com/OpenSourceMalaria/OSM_To_Do_List/issues/538) for my thesis and for the eventual paper on this work.

To get this started it would be super helpful if the entrants from the competition (@spadavec https://github.com/spadavec, @holeung https://github.com/holeung, @gcincilla https://github.com/gcincilla, @kellerberrin https://github.com/kellerberrin, @IamDavyG https://github.com/IamDavyG, @jon-c-silva) were able to provide a brief summary description about the models you developed for a broader audience to understand.

There is a Google Sheet here https://docs.google.com/spreadsheets/d/1pY6sYXIw66jnzUO3CoP8HceYdDjLRvwg5_pLkBY1Wek/edit#gid=0 where a new column could be added for this description (in addition to the Summary column already there).

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/OpenSourceMalaria/Series4/issues/70?email_source=notifications&email_token=ABXY3BMYQCP6R4OHYRFXSD3PX7U37A5CNFSM4HRF6IL2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4GWX4M3Q, or mute the thread https://github.com/notifications/unsubscribe-auth/ABXY3BKWFHVWWPSCFXS37Q3PX7U37ANCNFSM4HRF6ILQ .

-- Best Regards,

James

[edwintse]

Closing. Taken onto the predictive modelling repo here