v0.3.2

#210 Fix the use of deprecated built-in functions with DGL 1.0 (kudos to @eereenah-fast)

v0.3.1

• #189 Fix mask error for MoleculeCSVDataset (kudos to @nbrosse)
• #197 Unpin scikit-learn version (kudos to @flying-sheep)
• #202 Add GATv2 (kudos to @rgasper)
• #203 Add allow_zero_in_degree option for GAT and GCN (kudos to @astolman)

v0.3.0

This release:

• #168 Fix Misalignment in PDBBind
• #170 Fix Featurization for PDBBind
• #172 Change Default Number of Processes for rexgen_direct
• #177 Fix An Issue Related to JTVAE due to a new version of DGL
• #179 Update Version Requirement of scikit-learn
• #180 Provide module counterpart of some functional APIs, including MolToBigraph for mol_to_bigraph and SMILESToBigraph for smiles_to_bigraph

v0.2.9

This release:

• #149 Make RDKit an optional dependency
• #77 PotentialNet
• #154 Minor fix for MGCN
• #159 Multitask Molecular Property Prediction with GNN

v0.2.8

This release:

• #145 Fix a critical bug in scaffold split as reported in #144
• #146 Set bias in GAT to be True by default for compatibility with DGL 0.6.1

v0.2.7

This release:

• #109 Allow sanitizing molecules when removing hydrogens in data loading (kudos to @JoshuaMeyers)
• #114 Allow configuring normalization for GCN and GCNPredictor
• #116 #118 #128 #129 #131 #135 Fix for rexgen (kudos to @mar-volk)
• #120 #121 #122 Neural Fingerprint
• #110 Allow GNN pre-training for attribute masking and supervised learning (kudos to @wenx00)
• #133 Add timestep in early stopping
• #138 #139 PAGTN (kudos to @VIGNESHinZONE)
• #141 Fix for JTVAE

v0.2.6

This release:

• includes various pre-trained models for 12 datasets in MoleculeNet. See here
• fixes two bugs related to MoleculeCSVDataset and the WLN example for reaction prediction.
• supports two kinds of scaffold split, one based on MurckoDecompose, the other based on MurckoScaffoldSmiles

v0.2.5

This release supports DGL 0.5.0+ along with some feature enhancement. Note that it is recommended to use DGL 0.5.2+, which fixes some bugs. The feature enhancement includes:

• #67 Support for parallel graph construction in MoleculeCSVDataset (kudos to @sooheon )
• Better support for molecular property prediction on custom CSV datasets, which can be found here
• #65 #66 Bug fix for inference with fine-tuned GINs on custom CSV datasets (kudos to @skrsna )
• #73 Support for self loops in bond featurization
• #74 Support for virtual nodes in graph construction and featurization
• #75 Bug fix for featurizing self loops in graphs without real edges
• #61 #81 #82 #83 #86 Built-in support for more datasets from MoleculeNet
• #79 #80 Support for PR-AUC Metric
• #87 Filter out invalid molecules for MoleculeCSVDataset
• #84 #85 Support non-ring systems in scaffold split (kudos to @skrsna )