This repository hosts the codes for the TPN2.0 models, including the VAE-based and transformer-based ones. The code can be run on a shortlisted mock data provided in the mock_data folder. It will return the model performance based on the mock data, which should be close to 0 as the mock data were based on random distributions. On a normal computer, even without a GPU, this code with the small simulated dataset provided should not take longer than 10 minutes to run. The code serves only for demonstrating the algorithm architecture and training process. As the data is simulated from random distribution, the outputted pearson's R does not have a significant meaning.
- To run the VAE-based model
python main_VAE.py
- To run the transformer-based model
python transformer.py
| Parameter | Units | 95% CI Range (NICU) |
|---|---|---|
| gest_age | Days | 164-283 |
| bw | Kgs | 0.5-4.0 |
| day_since_birth | Days | 0-157 |
| RxDay | Days | 1-103 |
| TodaysWeight | Kgs | 0.65-5.80 |
| TPNHours | Hours | 24 |
| max_chole_TPNEHR | - | 1 if cholestasis |
| Alb_lab_value | g/dL | 1.7-3.7 |
| Ca_lab_value | mg/dL | 7.4-11.0 |
| Cl_lab_value | mEq/L | 92-113 |
| Glu_lab_value | mg/dL | 61-182 |
| Na_lab_value | mEq/L | 130-147 |
| BUN_lab_value | mg/dL | 6-57 |
| Cr_lab_value | mg/dL | 0.13-1.30 |
| Tri_lab_value | mg/dL | 22-247 |
| ALKP_lab_value | U/L | 64-731 |
| CaI_lab_value | mg/dL | 1.04-1.56 |
| CO2_lab_value | mEq/L | 18-38 |
| PO4_lab_value | mg/dL | 3.0-7.3 |
| K_lab_value | mEq/L | 2.6-5.8 |
| Mg_lab_value | mg/dL | 1.6-3.2 |
| AST_lab_value | U/L | 16-258 |
| ALT_lab_value | U/L | 6-224 |
| FluidDose | mL/kg/day | 80-170 |
| VTBI | mL | 55-626 |
| InfusionRate | mL/hr | 2.29-26.38 |
| FatInfusionRate | mL/hr | 0.00-3.03 |
| ProtocolName_NEONATAL | None | 1 if NEONATAL; 0 means PEDIATRIC |
| LineID_2 | None | 1 if peripheral; 0 means central |
| gender_concept_id_8507 | None | 1 if male |
| gender_concept_id_8532 | None | 1 if female |
| race_concept_id_8515 | None | 1 if Asian |
| race_concept_id_8516 | None | 1 if Black |
| race_concept_id_8527 | None | 1 if White |
| race_concept_id_8557 | None | 1 if Native Hawaiian or Other Pacific Islander |
| race_concept_id_8657 | None | 1 if American Indian or Alaska Native |
| FatProduct_SMOFlipid 20% | None | 1 if SMOFlipid 20% |
| FatProduct_Intralipid 20% | None | 1 if Intralipid 20% |
| FatProduct_Omegaven 10% | None | 1 if Omegaven 20% |
| Encoded_0 (0-32) | None | Encoded EHR representation from AE (32 dimensions) |