ailab-bio/PROTAC-Splitter-EncoderDecoder-lr_reduce-opt25-rand-smiles

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: 0.3251
• All Ligands Equal: 0.5455
• E3 Tanimoto Similarity: 0.0
• E3 Graph Edit Distance Norm: inf
• E3 Heavy Atoms Difference Norm: 0.0123
• Linker Tanimoto Similarity: 0.0
• Tanimoto Similarity: 0.0
• E3 Valid: 0.9866
• Linker Heavy Atoms Difference Norm: -0.0013
• Num Fragments: 2.9997
• Heavy Atoms Difference: 5.7446
• E3 Has Attachment Point(s): 0.9866
• E3 Equal: 0.8076
• Reassembly: 0.5548
• Poi Has Attachment Point(s): 0.9458
• Poi Graph Edit Distance Norm: inf
• Linker Graph Edit Distance: inf
• Poi Heavy Atoms Difference Norm: 0.0510
• Linker Has Attachment Point(s): 0.9952
• Poi Equal: 0.7632
• E3 Heavy Atoms Difference: 0.4690
• Has All Attachment Points: 0.9866
• Linker Graph Edit Distance Norm: inf
• Has Three Substructures: 0.9990
• Linker Equal: 0.7856
• Poi Valid: 0.9458
• Valid: 0.9308
• Poi Tanimoto Similarity: 0.0
• Reassembly Nostereo: 0.5789
• Linker Valid: 0.9952
• Heavy Atoms Difference Norm: 0.0744
• Linker Heavy Atoms Difference: 0.2002
• Poi Heavy Atoms Difference: 1.7548
• Poi Graph Edit Distance: inf
• E3 Graph Edit Distance: inf

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: 800
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	All Ligands Equal	E3 Tanimoto Similarity	E3 Graph Edit Distance Norm	E3 Heavy Atoms Difference Norm	Linker Tanimoto Similarity	Tanimoto Similarity	E3 Valid	Linker Heavy Atoms Difference Norm	Num Fragments	Heavy Atoms Difference	E3 Has Attachment Point(s)	E3 Equal	Reassembly	Poi Has Attachment Point(s)	Poi Graph Edit Distance Norm	Linker Graph Edit Distance	Poi Heavy Atoms Difference Norm	Linker Has Attachment Point(s)	Poi Equal	E3 Heavy Atoms Difference	Has All Attachment Points	Linker Graph Edit Distance Norm	Has Three Substructures	Linker Equal	Poi Valid	Valid	Poi Tanimoto Similarity	Reassembly Nostereo	Linker Valid	Heavy Atoms Difference Norm	Linker Heavy Atoms Difference	Poi Heavy Atoms Difference	Poi Graph Edit Distance	E3 Graph Edit Distance
0.0107	0.4932	5000	0.3008	0.5011	0.0	inf	0.0105	0.0	0.0	0.9859	0.0173	3.0002	4.6779	0.9859	0.7894	0.5097	0.9624	inf	inf	0.0317	0.9978	0.7466	0.4474	0.9880	inf	0.9993	0.7242	0.9624	0.9481	0.0	0.5343	0.9978	0.0617	0.5834	1.1262	inf	inf
0.0058	0.7398	7500	0.3187	0.5325	0.0	inf	0.0172	0.0	0.0	0.9835	0.0039	2.9999	4.7889	0.9835	0.8015	0.5402	0.9616	inf	59313031161473085686992099539504630098717768674496038276431872.0000	0.0264	0.9941	0.7586	0.6172	0.9898	inf	0.9992	0.7643	0.9616	0.9417	0.0	0.5643	0.9941	0.0626	0.2995	0.9886	inf	inf
0.0049	0.9864	10000	0.3251	0.5455	0.0	inf	0.0123	0.0	0.0	0.9866	-0.0013	2.9997	5.7446	0.9866	0.8076	0.5548	0.9458	inf	inf	0.0510	0.9952	0.7632	0.4690	0.9866	inf	0.9990	0.7856	0.9458	0.9308	0.0	0.5789	0.9952	0.0744	0.2002	1.7548	inf	inf

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.19.1