---
dataset_info:
  features:
  - name: vh
    dtype: string
  - name: vl
    dtype: string
  - name: label
    dtype:
      class_label:
        names:
          '0': non-binder
          '1': binder
  splits:
  - name: train
    num_bytes: 4428189
    num_examples: 18223
  - name: eval
    num_bytes: 553554
    num_examples: 2278
  - name: test
    num_bytes: 553554
    num_examples: 2278
  download_size: 415916
  dataset_size: 5535297
---
# HER2 binding dataset 

HER2 binding antibodies have been obtained from the [Github repo](https://github.com/dahjan/DMS_opt) for [Mason et al. (2021)](https://www.nature.com/articles/s41551-021-00699-9).

Labels for antibody sequences were generated using scripts in the above Github repo. The number of negatives and positives were balanced through random undersampling using [imbalanced-learn](https://github.com/scikit-learn-contrib/imbalanced-learn), and sequences were deduplicated.

The dataset has:
* 39108 antibodies in total
* 22779 antibodies after undersampling and deduplication
* 18223 in the training set, 2278 in the evaluation set, and 2278 in the test set.

Example

| vh_full | vl_full | label |
| ------- | ------- | ----- |
| EVQ...  | DIQ...  | 1     |
| EVQ...  | DIQ...  | 0     |
| EVQ...  | DIQ...  | 0     |
| EVQ...  | DIQ...  | 1     |

References

* [Mason et al. (2021) paper](https://www.nature.com/articles/s41551-021-00699-9)