Priyadarshini Rai, Atishay Jain, Shivani Kumar, Neha Jha, Divya Sharma, Smriti Chawla, Abhijit Raj, Apoorva Gupta, Sarita Poonia, Angshul Majumdar, Tanmoy Chakraborty, Gaurav Ahuja, and Debarka Sengupta

Pathomap provides a strategy to obtain an unbiased continuous representation of disease causing genes and their tissue specificity.

https://drive.google.com/drive/folders/1GJE6zOrQYkavz0veiW_6SPv6uo6v4-ja?usp=sharing

https://drive.google.com/file/d/19H4QagNJTDzQCRVBOWn8oaPGG4FKsFJU/view?usp=sharing

Classification.ipynb and Supplementary File 1.csv

https://drive.google.com/drive/folders/1Hrdw_P2umP81SzrODAvrZhLKu2OwKDC_?usp=sharing

1. Classification.ipynb: Script used to perform classification.
2. Function to Get Embeddings.py: To get embedding(s) of a word(s) using PathoBERT.
3. GeneDisFreq.R: Function to get PMID(s) in which a gene-disease pair is present.
4. Supplementary File 1.csv: Pathological and non-pathological abstracts used as ground truth for the classification task.
5. Supplementary File 2.csv: DisGeNET gene-disease pairs cosine similarity using six different models, namely, PathoBERT, Word2vec, BioBERT, BioSentVec, BERT Human, and Word2vec Human.
6. Supplementary File 3.csv: Random gene-disease pairs cosine similarity using six different models, namely, PathoBERT, Word2vec, BioBERT, BioSentVec, BERT Human, and Word2vec Human.

To get PathoBERT embeddings of suppose diseases cardiac homeostasis and rhythm disorder, we will use function get_doc_vector written inside script Function to Get Embeddings.py.

E1 = get_doc_vector('cardiac homeostasis', tokenizer, model)
E2 = get_doc_vector('rhythm disorder', tokenizer, model)

To compute the cosine similarity between the embeddings of diseases cardiac homeostasis and rhythm disorder, we can use the function cosine_similarity

CS = cosine_similarity([E1.numpy()], [E2.numpy()])

F = GeneDisFreq('BRCA1', breast cancer')
F = GeneDisFreq('MASP1', 'cardiac homeostasis')