Project Website: https://bios.idea.edu.cn
BIOS: An Algorithmically Generated Biomedical Knowledge Graph Paper
This repo is under construction.
Software dependencies are listed in requirements.txt. At least one GPU is needed for training.
Just clone the github repo.
Use following command to train model.
CUDA_VISIBLE_DEVICES=0 python main.py --bag_size 16 --aggr one --batch_size 8 --gradient_accumulation_steps 2 --data_path ./data/1224 --output_base_dir ./output_1224 --limit_dis '3,10'
Use following command to inference input_data.
CUDA_VISIBLE_DEVICES=0 python predict.py input_data output_1224/entity_cls_False_0.1_one_binary_2_2e-05_16_16_0.0_3,10_coder/output.txt
input_data is data with json lines, here is a example:
{"text": "9-(s)-(2,3-dihydroxypropyl)adenine inhibits the transformation of chick embryo fibroblasts infected with rous sarcoma virus: evidence for inhibition of enzymatic activity of isolated cellular protein kinases by the drug..", "h": {"pos": [66, 78], "id": "CN00468812", "name": "chick embryo"}, "t": {"pos": [105, 123], "id": "CN00449160", "name": "rous sarcoma virus"}, "distance": 5}
{"text": "9-(s)-(2,3-dihydroxypropyl)adenine inhibits the transformation of chick embryo fibroblasts infected with rous sarcoma virus: evidence for inhibition of enzymatic activity of isolated cellular protein kinases by the drug..", "h": {"pos": [66, 78], "id": "CN00468812", "name": "chick embryo"}, "t": {"pos": [138, 148], "id": "CN00129331", "name": "inhibition"}, "distance": 10}
{"text": "9-(s)-(2,3-dihydroxypropyl)adenine inhibits the transformation of chick embryo fibroblasts infected with rous sarcoma virus: evidence for inhibition of enzymatic activity of isolated cellular protein kinases by the drug..", "h": {"pos": [79, 90], "id": "CN00453821", "name": "fibroblasts"}, "t": {"pos": [105, 123], "id": "CN00449160", "name": "rous sarcoma virus"}, "distance": 4}
{"text": "9-(s)-(2,3-dihydroxypropyl)adenine inhibits the transformation of chick embryo fibroblasts infected with rous sarcoma virus: evidence for inhibition of enzymatic activity of isolated cellular protein kinases by the drug..", "h": {"pos": [79, 90], "id": "CN00453821", "name": "fibroblasts"}, "t": {"pos": [138, 148], "id": "CN00129331", "name": "inhibition"}, "distance": 9}
Inference with CPU can be slow, a GPU is needed for fast inference.
Example output file can be:
{"text": "[CLS] sublethal irradiation with ultraviolet b light appeared to diminish ebv antigen expression ( gp350 / 220 ) during the first 48 to 72 hours in culture , whereas there was no change in the expression of mhc class i or [UNK] immunoglobulin [UNK] host cell [UNK] proteins [UNK] , and an apparent increase in the expression of host cell autoantigens . [SEP]", "h": "immunoglobulin", "t": "proteins", "predict_rel": ["is a"]}
{"text": "[CLS] interestingly , this peptide was recognized by only a few ( less than or equal to 7 % ) of sle sera , while 63 % of pss sera and 46 % of sle sera tested in parallel [UNK] possessed [UNK] antibodies reacting in elisa with purified 60 - kd [UNK] ssa protein [UNK] . [SEP]", "h": "possessed", "t": "ssa protein", "predict_rel": ["inverse involved in"]}