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: 0.3420
Num Fragments: 3.0002
Linker Heavy Atoms Difference: 0.1689
Linker Graph Edit Distance: 37181303116147309234962303999400365269286085344573508414341120.0000
Tanimoto Similarity: 0.0
Linker Tanimoto Similarity: 0.0
E3 Valid: 0.9732
Linker Has Attachment Point(s): 0.9963
Poi Equal: 0.7897
Heavy Atoms Difference: 8.0244
Poi Has Attachment Point(s): 0.9305
E3 Equal: 0.8302
Linker Graph Edit Distance Norm: inf
E3 Has Attachment Point(s): 0.9732
Has Three Substructures: 0.9995
Poi Heavy Atoms Difference Norm: 0.0690
Linker Equal: 0.8419
Heavy Atoms Difference Norm: 0.1076
E3 Heavy Atoms Difference: 1.0454
Poi Valid: 0.9305
Valid: 0.9027
Linker Heavy Atoms Difference Norm: -0.0046
Has All Attachment Points: 0.9796
E3 Graph Edit Distance: inf
Linker Valid: 0.9963
Poi Tanimoto Similarity: 0.0
Poi Graph Edit Distance Norm: inf
Poi Heavy Atoms Difference: 2.0482
Poi Graph Edit Distance: inf
Reassembly Nostereo: 0.6261
E3 Graph Edit Distance Norm: inf
E3 Heavy Atoms Difference Norm: 0.0335
Reassembly: 0.6073
All Ligands Equal: 0.5992
E3 Tanimoto Similarity: 0.0

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
training_steps: 100000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Num Fragments	Linker Heavy Atoms Difference	Linker Graph Edit Distance	Tanimoto Similarity	Linker Tanimoto Similarity	E3 Valid	Linker Has Attachment Point(s)	Poi Equal	Heavy Atoms Difference	Poi Has Attachment Point(s)	E3 Equal	Linker Graph Edit Distance Norm	E3 Has Attachment Point(s)	Has Three Substructures	Poi Heavy Atoms Difference Norm	Linker Equal	Heavy Atoms Difference Norm	E3 Heavy Atoms Difference	Poi Valid	Valid	Linker Heavy Atoms Difference Norm	Has All Attachment Points	E3 Graph Edit Distance	Linker Valid	Poi Tanimoto Similarity	Poi Graph Edit Distance Norm	Poi Heavy Atoms Difference	Poi Graph Edit Distance	Reassembly Nostereo	E3 Graph Edit Distance Norm	E3 Heavy Atoms Difference Norm	Reassembly	All Ligands Equal	E3 Tanimoto Similarity
0.0005	7.8911	80000	0.3375	2.9994	0.2223	inf	0.0	0.0	0.9717	0.9961	0.7867	9.0605	0.9179	0.8268	inf	0.9717	0.9994	0.0813	0.8393	0.1211	0.9062	0.9179	0.8893	0.0019	0.9788	inf	0.9961	0.0	inf	2.4517	820644475920679939503384490879847350906393395732940812913213440.0000	0.6211	inf	0.0334	0.6011	0.5936	0.0
0.0005	9.8639	100000	0.3420	3.0002	0.1689	37181303116147309234962303999400365269286085344573508414341120.0000	0.0	0.0	0.9732	0.9963	0.7897	8.0244	0.9305	0.8302	inf	0.9732	0.9995	0.0690	0.8419	0.1076	1.0454	0.9305	0.9027	-0.0046	0.9796	inf	0.9963	0.0	inf	2.0482	inf	0.6261	inf	0.0335	0.6073	0.5992	0.0

Framework versions

Transformers 4.44.2
Pytorch 2.4.1+cu121
Datasets 3.0.0
Tokenizers 0.19.1