Update README.md
Browse files
README.md
CHANGED
|
@@ -36,4 +36,23 @@ DistilProtBert model was pretrained on [Uniref50](https://www.uniprot.org/downlo
|
|
| 36 |
# Pretraining procedure
|
| 37 |
|
| 38 |
Preprocessing was done using ProtBert's tokenizer.
|
| 39 |
-
The details of the masking procedure for each sequence followed the original Bert (as mentioned in [ProtBert](https://huggingface.co/Rostlab/prot_bert)).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
# Pretraining procedure
|
| 37 |
|
| 38 |
Preprocessing was done using ProtBert's tokenizer.
|
| 39 |
+
The details of the masking procedure for each sequence followed the original Bert (as mentioned in [ProtBert](https://huggingface.co/Rostlab/prot_bert)).
|
| 40 |
+
|
| 41 |
+
The model was pretrained on a single DGX cluster 3 epochs in total. local batch size was 16, the optimizer used was AdamW with a learning rate of 5e-5 and mixed precision settings.
|
| 42 |
+
|
| 43 |
+
## Evaluation results
|
| 44 |
+
|
| 45 |
+
When fine-tuned on downstream tasks, this model achieves the following results:
|
| 46 |
+
|
| 47 |
+
Test results :
|
| 48 |
+
|
| 49 |
+
| Task/Dataset | secondary structure (3-states) | Membrane |
|
| 50 |
+
|:-----:|:-----:|:-----:|:-----:|
|
| 51 |
+
| CASP12 | 72 | |
|
| 52 |
+
| TS115 | 81 | |
|
| 53 |
+
| CB513 | 79 | |
|
| 54 |
+
| DeepLoc | | 86 |
|
| 55 |
+
|
| 56 |
+
Distinguish between:
|
| 57 |
+
|
| 58 |
+
### BibTeX entry and citation info
|