How can we enhance the final results to obtain more meaningful and insightful biological findings?

[PhoebeMagdy]

During the benchmarking process of various models generated by executing a bash script designed to explore different deep learning architectures and combinations of omic data modalities, the different models were evaluated based on their Pearson correlation scores.

It was observed that the highest correlation score achieved was only 0.389583, indicating a relatively low correlation coefficient. Furthermore, upon extracting the top 10 important features, none of them were found to be directly associated with "Erlotinib" in the existing literature, this indeed was expected because of the low correlation

• Could you please begin by explaining which two numerical variables were used to calculate the Pearson correlation coefficient in each model?
• What factors may have influenced the unfavorable outcome?
• How can we enhance the results to obtain more meaningful and insightful biological findings?

[borauyar]

• You compute the correlation between known and predicted values of Erlotinib response in the test dataset.
• The outcome isn't that unfavorable. The drug response values between identical cell lines between different datasets might not be perfectly correlated. What your model can predict is first of all bounded by the reproducibility of drug response scores of the same drug in the two different datasets. This also depends on the kind of drug. For Erlotinib, a correlation of 0.4 is not that bad.
• You can enhance the results by running larger models for longer HPO training steps using more input features. The top markers you found may not necessarily be wrong, but could be indirectly associated with the actual markers. However, I see that with better performance models, you usually find the known targets of Erlotinib with a longer training regimen with a larger feature space.

[PhoebeMagdy]

Thank you for the illustration and for all the effort you put into this workshop which was highly informative and greatly appreciated.

[borauyar]

Thank you!