This model is a fine-tuned version of seyonec/ChemBERTa-zinc-base-v1 on the ailab-bio/PROTAC-Splitter-Dataset dataset. It achieves the following results on the evaluation set:

Loss: 0.3751
Reassembly: 0.5970
E3 Graph Edit Distance Norm: inf
Linker Graph Edit Distance Norm: inf
All Ligands Equal: 0.5899
Poi Heavy Atoms Difference Norm: 0.0419
Poi Graph Edit Distance Norm: inf
Linker Heavy Atoms Difference: 0.2883
Linker Equal: 0.8472
Poi Has Attachment Point(s): 0.9510
E3 Tanimoto Similarity: 0.0
Reassembly Nostereo: 0.6329
E3 Graph Edit Distance: inf
Linker Has Attachment Point(s): 0.9977
E3 Heavy Atoms Difference Norm: 0.0037
Linker Valid: 0.9977
Poi Equal: 0.7890
Linker Graph Edit Distance: 23016997167138810478786188190104988023843767118069314732687360.0000
E3 Equal: 0.8240
Num Fragments: 3.0003
E3 Heavy Atoms Difference: 0.2737
Poi Graph Edit Distance: inf
Has Three Substructures: 0.9996
Heavy Atoms Difference: 4.7685
Linker Tanimoto Similarity: 0.0
Poi Heavy Atoms Difference: 1.3950
E3 Valid: 0.9944
Poi Tanimoto Similarity: 0.0
Valid: 0.9450
Heavy Atoms Difference Norm: 0.0634
Tanimoto Similarity: 0.0
E3 Has Attachment Point(s): 0.9944
Has All Attachment Points: 0.9938
Poi Valid: 0.9510
Linker Heavy Atoms Difference Norm: 0.0067

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-05
train_batch_size: 128
eval_batch_size: 64
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: reduce_lr_on_plateau
lr_scheduler_warmup_steps: 400
training_steps: 100000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Reassembly	E3 Graph Edit Distance Norm	Linker Graph Edit Distance Norm	All Ligands Equal	Poi Heavy Atoms Difference Norm	Poi Graph Edit Distance Norm	Linker Heavy Atoms Difference	Linker Equal	Poi Has Attachment Point(s)	E3 Tanimoto Similarity	Reassembly Nostereo	E3 Graph Edit Distance	Linker Has Attachment Point(s)	E3 Heavy Atoms Difference Norm	Linker Valid	Poi Equal	Linker Graph Edit Distance	E3 Equal	Num Fragments	E3 Heavy Atoms Difference	Poi Graph Edit Distance	Has Three Substructures	Heavy Atoms Difference	Linker Tanimoto Similarity	Poi Heavy Atoms Difference	E3 Valid	Poi Tanimoto Similarity	Valid	Heavy Atoms Difference Norm	Tanimoto Similarity	E3 Has Attachment Point(s)	Has All Attachment Points	Poi Valid	Linker Heavy Atoms Difference Norm
0.0006	15.7822	80000	0.3655	0.5977	inf	inf	0.5904	0.0477	inf	0.1988	0.8371	0.9528	0.0	0.6278	inf	0.9952	0.0221	0.9952	0.7851	inf	0.8250	2.9996	0.7440	inf	0.9994	5.9070	0.0	1.4796	0.9796	0.0	0.9315	0.0782	0.0	0.9796	0.9879	0.9528	-0.0014
0.0006	19.7278	100000	0.3751	0.5970	inf	inf	0.5899	0.0419	inf	0.2883	0.8472	0.9510	0.0	0.6329	inf	0.9977	0.0037	0.9977	0.7890	23016997167138810478786188190104988023843767118069314732687360.0000	0.8240	3.0003	0.2737	inf	0.9996	4.7685	0.0	1.3950	0.9944	0.0	0.9450	0.0634	0.0	0.9944	0.9938	0.9510	0.0067

Framework versions

Transformers 4.44.2
Pytorch 2.4.1+cu121
Datasets 3.0.0
Tokenizers 0.19.1