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: 0.3086
Poi Heavy Atoms Difference: 2.1208
E3 Valid: 0.9896
Poi Valid: 0.9272
Poi Has Attachment Point(s): 0.9272
All Ligands Equal: 0.5462
Valid: 0.9157
Reassembly: 0.5544
Poi Tanimoto Similarity: 0.0
Linker Tanimoto Similarity: 0.0
Poi Graph Edit Distance: inf
Linker Heavy Atoms Difference: 0.3144
Linker Graph Edit Distance Norm: inf
E3 Graph Edit Distance Norm: inf
Num Fragments: 2.9998
E3 Heavy Atoms Difference Norm: 0.0131
Linker Valid: 0.9961
E3 Heavy Atoms Difference: 0.5553
E3 Tanimoto Similarity: 0.0
Poi Heavy Atoms Difference Norm: 0.0719
Reassembly Nostereo: 0.5796
Linker Equal: 0.7666
Linker Has Attachment Point(s): 0.9961
Has All Attachment Points: 0.9836
Poi Equal: 0.7680
E3 Graph Edit Distance: inf
Tanimoto Similarity: 0.0
E3 Has Attachment Point(s): 0.9896
Poi Graph Edit Distance Norm: inf
E3 Equal: 0.8045
Heavy Atoms Difference Norm: 0.0939
Has Three Substructures: 0.9991
Linker Graph Edit Distance: inf
Heavy Atoms Difference: 7.0102
Linker Heavy Atoms Difference Norm: 0.0033

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: cosine_with_restarts
lr_scheduler_warmup_steps: 100
training_steps: 10000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Poi Heavy Atoms Difference	E3 Valid	Poi Valid	Poi Has Attachment Point(s)	All Ligands Equal	Valid	Reassembly	Poi Graph Edit Distance	Linker Heavy Atoms Difference	Linker Graph Edit Distance Norm	E3 Graph Edit Distance Norm	Num Fragments	E3 Heavy Atoms Difference Norm	Linker Valid	E3 Heavy Atoms Difference	Poi Heavy Atoms Difference Norm	Reassembly Nostereo	Linker Equal	Linker Has Attachment Point(s)	Has All Attachment Points	Poi Equal	E3 Graph Edit Distance	E3 Has Attachment Point(s)	Poi Graph Edit Distance Norm	E3 Equal	Heavy Atoms Difference Norm	Has Three Substructures	Linker Graph Edit Distance	Heavy Atoms Difference	Linker Heavy Atoms Difference Norm
0.0156	0.4932	5000	0.2891	2.3129	0.9916	0.9252	0.9252	0.4596	0.9169	0.4665	inf	0.3805	inf	inf	3.0006	0.0061	0.9965	0.3913	0.0739	0.4911	0.6623	0.9965	0.9863	0.7265	inf	0.9916	inf	0.7815	0.0918	0.9988	35410764872521246890440289523238443113605795566260484204134400.0000	7.0030	0.0005
0.0077	0.7398	7500	0.3073	2.5750	0.9857	0.9161	0.9161	0.4898	0.8987	0.4975	inf	0.4483	inf	inf	3.0002	0.0088	0.9943	0.4845	0.0832	0.5220	0.7095	0.9943	0.9812	0.7363	inf	0.9857	inf	0.7908	0.1093	0.9995	56657223796033995024704463237181508981769272906016774726615040.0000	8.2839	0.0090
0.0046	0.9864	10000	0.3086	2.1208	0.9896	0.9272	0.9272	0.5462	0.9157	0.5544	inf	0.3144	inf	inf	2.9998	0.0131	0.9961	0.5553	0.0719	0.5796	0.7666	0.9961	0.9836	0.7680	inf	0.9896	inf	0.8045	0.0939	0.9991	inf	7.0102	0.0033

Framework versions

Transformers 4.44.2
Pytorch 2.4.1+cu121
Datasets 3.0.0
Tokenizers 0.19.1