For both Task 1 and Task 2 of the DREAM Olfaction Prediction Challenge, we submitted baseline predictions to the leaderboard as **Baseline_Model_Task1LB** and **Baseline_Model_Task2LB**, respectively.
**Task 1 (Leaderboard baseline)**
All molecules in the leaderboard and test set are present at a different concentration in the training set. The baseline model does not account for concentration effects—it simply copied the RATA data from the wrong concentration of the same molecule. It achieved a Pearson correlation of **0.501 **and Cosine distance of **0.359 **on the leaderboard set.
**Task 2 (Leaderboard baseline)**
For each mixture in the leaderboard set, we first identified its components based on the provided Stimulus definition. We then located the corresponding molecules in the Task 2 training set and extracted their RATA data. These RATA data were then averaged (simple unweighted mean) and used as the predicted perceptual data for the mixture. The baseline model achieved a Pearson correlation of **0.653** and Cosine distance of **0.22** on the leaderboard set. Note that the concentrations of the individual molecules in the training data do not necessarily match their concentrations in the mixtures.
Created by Xuebo Song Songxuebo @Songxuebo I tried to submit the write-up by it was invalid because its not a folder its a file, but folder can not be submitted because "submission to challenge" option does not appear. how can i fix it?
In the baseline model, we used the 51-dimensional RATA vector (semantic descriptor ratings) as input features, rather than molecular embeddings or fingerprints. Kindly let me know what was the dimension of embeddings used in baseline. The paper mentioned 256 but openpom has 512 D possible as well. I am really excited to be a part of this competition but I need aome guidance and I hooe the approach would be liked by the community. We didn’t incorporate the 512-D embeddings from OpenPOM in the current baseline, but it’s an excellent suggestion. We're very much looking forward to your results! @Songxuebo have you used 512-D Embeddings from SMILES Using OpenPOM for baseline? Hi, here is the link for [OpenPOM](https://github.com/BioMachineLearning/openpom). Hi dskhanirfan, We’ve described this in the baseline write-up, but unfortunately, we’re not able to share the related code before the challenge concludes. Thanks for your understanding! @jeriscience can you kindly upload the POM file as well? Is the baseline code available? the intensity column in task1 train quantifies the concentration effects? @Songxuebo
Drop files to upload
Baseline Write-up for DREAM Olfaction Prediction Challenge page is loading…