ideasbyjin
commited on
Commit
•
445484b
1
Parent(s):
b3a60bb
Title
Browse files- LICENSE.md +1 -1
LICENSE.md
CHANGED
@@ -1,4 +1,4 @@
|
|
1 |
-
AntiBERTa2 License
|
2 |
Version 1.0, December 2023 [Alchemab](http://alchemab.com)
|
3 |
|
4 |
AntiBERTa2 models are antibody-specific Large Language Models (LLM) developed by Alchemab Therapeutics Ltd. Currently AntiBERTa2 includes the following two architectures referred to as AntiBERTa2 (220 M parameters) and AntiBERTa2-CSSP (220 M parameters). Based on the RoFormer LLM from natural language processing, AntiBERTa2 is pre-trained using MLM on 779.4 million unpaired and paired antibody sequences. AntiBERTa2-CSSP refers to a multimodal version of the AntiBERTa2 LLM that leverages the contrastive pre-training objective from the CLIP method; over 1000 antibody crystal structures from the Protein Data Bank were used. After pre-training, AntiBERTa2 develops a rich representation of antibody sequences that can be leveraged for an extensive range of tasks, including (without limitation) antigen binding prediction.
|
|
|
1 |
+
# AntiBERTa2 License
|
2 |
Version 1.0, December 2023 [Alchemab](http://alchemab.com)
|
3 |
|
4 |
AntiBERTa2 models are antibody-specific Large Language Models (LLM) developed by Alchemab Therapeutics Ltd. Currently AntiBERTa2 includes the following two architectures referred to as AntiBERTa2 (220 M parameters) and AntiBERTa2-CSSP (220 M parameters). Based on the RoFormer LLM from natural language processing, AntiBERTa2 is pre-trained using MLM on 779.4 million unpaired and paired antibody sequences. AntiBERTa2-CSSP refers to a multimodal version of the AntiBERTa2 LLM that leverages the contrastive pre-training objective from the CLIP method; over 1000 antibody crystal structures from the Protein Data Bank were used. After pre-training, AntiBERTa2 develops a rich representation of antibody sequences that can be leveraged for an extensive range of tasks, including (without limitation) antigen binding prediction.
|