Comprehensive modeling
Create multiple models simultaneously
The comprehensive modeling feature allows you to simultaneously run multiple models
with different machine learning methods, molecular descriptors and validation protocols.
Please note that running multiple models may require significant computational resources and time.
Select the methods you want to use for the modeling:
Show advanced options>>
Considering the selection above, 120 models are requested to be created. Notice that only maximum 25 models based on descriptors can be started simultaneously.
Please note that running multiple models may require significant computational resources and time.
Select the training and validation sets:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
[...]
using unit:
Select the methods you want to use for the modeling:
Method[all] [none]
ASNN
Supersab, 1000 iterations, 3 neurons ensemble=64 additional param PARTITION=3,SELECTION=2
KNN
Max KNN = 100, Euclidian distance
LibSVM
classic algorithm, RBF
scale X: RANGE
FSMLR
MLRA
PLS
0 latent variables
WEKA-RF (classification only)
trees=10, features=0, depth=0
WEKA-J48 (classification only)
confidence = 0.25, inst per leaf = 2
XGBOOST
trees: 1024 with max depth of 8 eta=0.2
LSSVMG (GPU)
kernel:rbf cv:5
scale X: STANDARDIZE
scale Y: STANDARDIZE
CNF (no descr, MTL, GPU)
FPDim:32 nLayers:8 FFNET_dim:128,64 nnf_type=v2.2 batch=32 learning rate=0.001 CHAR
dropout=0.2 3D epochs=1000 early=0.1 activation=relu 10/10
scale X: STANDARDIZE
scale Y: STANDARDIZE
EAGCNG (no descr, MTL, GPU)
epochs=1000 norm=LAYER gate=false batch=512 dropout=0.2 learningRate=0.01 weightDecay=1.0E-4
method=concate n_sgc1=30,15,15,15,15 n_sgc2=60,30,30,30,30 n_den=64,32,12
CHEMPROP (no descr, GPU)
depth: 3 epoch: 1000 hidden=300 batch =64
scale X: STANDARDIZE
scale Y: STANDARDIZE
KGCNN ChemProp
batch=32 epochs=200 sn
scale X: STANDARDIZE
scale Y: STANDARDIZE
KGCNN AttFP
batch=32 epochs=200 sn
scale X: STANDARDIZE
scale Y: STANDARDIZE
Transformer CNN (no descr, MTL, GPU)
batch= 64 early=0.1 fixedrate R learning_rate=1.0E-4 shuffled epochs=100 shuffle=
early=0.1 10/10
scale X: STANDARDIZE
scale Y: STANDARDIZE
AttFP
batch=64 epochs=200
scale X: STANDARDIZE
scale Y: STANDARDIZE
ChemProp
depth: 5 epoch: 200 hidden=200 batch =64
scale X: STANDARDIZE
scale Y: STANDARDIZE
DNN
scale X: STANDARDIZE
scale Y: STANDARDIZE DNN: dense7new epochs: 1000 optimizer: adam batch: 1024 activation:
relu
RFR
trees=512
KGCNN DimeNetPP
batch=32 epochs=200
scale X: STANDARDIZE
scale Y: STANDARDIZE
KGCNN HamNet
batch=32 epochs=200
scale X: STANDARDIZE
scale Y: STANDARDIZE
KGCNN GIN
batch=32 epochs=200
scale X: STANDARDIZE
scale Y: STANDARDIZE
KGCNN GINE
batch=32 epochs=200
scale X: STANDARDIZE
scale Y: STANDARDIZE
KGCNN Schnet
batch=32 epochs=200
scale X: STANDARDIZE
scale Y: STANDARDIZE
KGCNN GAT
batch=32 epochs=200
scale X: STANDARDIZE
scale Y: STANDARDIZE
KGCNN GraphSAGE
batch=32 epochs=200
scale X: STANDARDIZE
scale Y: STANDARDIZE
KPLS
scale X: STANDARDIZE
scale Y: STANDARDIZE latent_variables = 0 lambda
DEEP
scale X: STANDARDIZE
scale Y: STANDARDIZE
+add a custom template |
Descriptors[all] [none]
CDK (3D)
[CDK2 (cons,topol,geom,elect,hybr)]
OEstate and ALogPS
[ALogPS, OEstate]
ISIDA Fragments (Length 2 - 4)
[Fragmentor (length: 2-4)]
Inductive Descriptors (3D)
[InductiveDescriptors]
Spectrophores (3D)
[Spectrophores (accuracy=20)]
QNPR (length 1 - 3)
[QNPR (length:1 - 3)]
Extended Functional Groups (EFG)
[StructuralAlerts]
SIRMS
[SIRMS (labels: charge)]
MOLD2
[MOLD2]
JPlogP
[JPlogP]
MAP4
[MAP4]
MORDRED (3D)
[MORDRED ( All)]
EPA T.E.S.T.
[EPA (T.E.S.T.)]
PyDescriptor (3D)
[PyDescriptor]
CDDD
[CDDD]
RDKIT (selected)
[RDKIT (3D blocks: 1-11 15-16)]
no descr (representation learning)
Correl. limit: 0.95 Variance threshold: 0.01, Maximum value: 999999,
No validation
OESTATE all
[OEstate]
+add a custom template |
Descriptor selection[all] [none]
Unsupervised forward selection
Correl. limit: 0.95 Variance threshold: 0.01, Maximum value: 999999, using UFS
No validation
Pairwise decorrelation (r < 0.95)
Correl. limit: 0.95 Variance threshold: 0.01, Maximum value: 999999,
No validation
+add a custom template |
Model validation[all] [none]
5-fold cross-validation
5-fold cross-validation
5-fold cross-validation (stratified - classification only)
5-fold cross-validation (stratified)
Bagging with 64 models
Bagging with 64 sets
Bagging with 64 models (stratified - classification only)
Bagging with 64 sets (stratified)
No validation (not recommended)
No validation
10 CV
10-fold cross-validation
+add a custom template |
Skip duplicates (if a model with the same configuration already exists)
Records with ranges
Show mixture options (required if "Remove salts" is not checked)>>
Show conditions>>
Conditions of experiments
Override default normalisation options
Save models
Enforce Single Task Learning (STL) for MTL method
Use Feature Net learning for multitask basket
Use sanitisation
(Experimental) Merge with the uploaded descriptor type:
(Experimental for sparse data) Substitute implicit values (separated by ;) with:
(Experimental for taxonomy) Use taxonomy:
Records with ranges
Include following records:
Include "approximately equals" records
Include interval records
Include "greater" and "less" records
Handling of records with ranges:
Use average values (for intervals) and boundary values (for greater and less ranges)
Handle ranges as ranges (experimental)
Include "approximately equals" records
Include interval records
Include "greater" and "less" records
Handling of records with ranges:
Use average values (for intervals) and boundary values (for greater and less ranges)
Handle ranges as ranges (experimental)
Show mixture options (required if "Remove salts" is not checked)>>
Mixture processing options:
| Additional validation option for mixtures (salts): |
|
Validation by each mixture component Validation by anions (ILs only) Validation by cations (ILs only) Validation by mixtures (each mixture is considered as a compound - likely to overfit) |
Show conditions>>
Conditions of experiments
Select a tool to optimize molecule structures
Override default normalisation options
| Descriptors normalization | |
| Values normalization |
Save models
Enforce Single Task Learning (STL) for MTL method
Use Feature Net learning for multitask basket
Use sanitisation
(Experimental) Merge with the uploaded descriptor type:
(Experimental for sparse data) Substitute implicit values (separated by ;) with:
(Experimental for taxonomy) Use taxonomy:
Considering the selection above, 120 models are requested to be created. Notice that only maximum 25 models based on descriptors can be started simultaneously.
Success!
You can track the status of your models in the