0.25
0.5
0.75
1.25
1.5
1.75
2
Meta-QSAR and Multi-Task QSAR Learning
Published on Jun 28, 2019178 Views
Larisa will present the results of the meta-QSAR project funded by EPSRC (Engineering and Physical Sciences Research Council UK) (‘learning to learn how to design drugs’ EP/K030469/1, EP/K030582/1). A
Related categories
Chapter list
Meta-qsar and multi-task qsar learning00:00
Outline of the talk00:29
Part I: meta-qsar01:15
Motivation01:18
Quantitative structure activity relationship (qsar)03:14
Qsar learning -‘no free lunch’04:21
The rational for meta-qsar learning05:10
Meta-learning approach06:12
Meta-learning task07:30
Baseline qsar learning: algorithms09:19
Baseline qsar learning: datasets - 110:05
Baseline qsar learning: datasets - 211:01
Chembl’sclassification of drug targets12:11
Molecular descriptors13:58
Representations15:48
Baseline qsar experiments16:42
Dataset representations17:41
Baseline qsar experiments with representations18:23
Meta-features for meta-qsar learning19:01
Meta-qsar ontology19:37
Dataset meta-features20:00
Drug target meta-features20:35
Drug target groupings21:33
The importance of each meta-feature in the classification task22:02
Meta-qsar dataset22:33
Meta-learning pipeline23:50
A meta-learning classification and ranking25:07
Ranking models25:45
Meta-qsar performance26:31
Meta-qsar: conclusion27:22
Part II: multi-task qsar learning39:29
The problem39:34
Multiple task learning (mtl)40:36
Types of mtl41:43
Baseline: single task learning (stl)43:10
Stl implementation43:50
Feature-based mtl (setting 1)44:22
The similarity of drug targets45:09
Results for L5 target classes46:58
Boxplot of rmse values47:46
Instance-based mtl (setting 2)47:58
Conclusions48:54
Availability49:46
Acknowledgements50:18
Thank you50:54