tide-var

This repository includes the python code of TIDE-Var and of T-ResNet, two deep neural models for the prediction of genetic variants in Mendelian diseases.

TIDE-Var is an implicit ensemble of deep neural networks that partially share parameters and in parallel contextually optimize the same objective function, borrowed from the recently proposed TabM model (Gorishniy et al., ICLR, 2025).

T-ResNet (Tabular Residual Neural Network) adopts a modular architecture with residual connections to support efficient hyperparameter optimization, along with a mini-batch balancing strategy to address class imbalance.

Datasets used for training the model are constiduted by 26-dimensional genetic, epigenetic and conservation features originally collected by Smedley et al, Am. J. Hum. Genet. 99, 595–606 (2016). Dataset splits used in this study are avaialble in src/tremm/data/tensors/<dataset_folder>/<dataset_name>/.

Install

python -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt
python -m pip install -e .

This installs the tremm CLI.

Data and datasets

The CLI downloads raw data into src/tremm/data/raw_data on first use.

tremm create_kfold --dataset_size 100000 --ext_test_ratio 0.2 --int_valid_ratio 0.2 --random_seed 42
tremm create_dataset --dataset_size 100000 --split_ratios 0.7 0.15 0.15 --positive_ratios 0.5 0.25 0.25 --random_seed 42
tremm create_n_kfold --dataset_size 100000 --ext_test_ratio 0.2 --global_valid_ratio 0.1 --num_internal_folds 5 --int_valid_ratio 0.2 --random_seed 42
tremm list

If you do not want to install the package:

PYTHONPATH=src python -m tremm.scripts.cli <subcommand> [args...]

Train

TabM (uses CUDA/MPS if available, otherwise CPU):

python src/train_tide_var.py -d kfold_100k -n <dataset_name> -e 25 -k 1 -m ple -de 12

Single-fold / single-worker run:

python src/train_tide_var.py -d kfold_100k -n <dataset_name> -e 25 -k 1 -m ple -de 12 --fold 0 --workers 1

Results are saved as CSVs under src/csv_results/.

ResNet:

1. Update DATASET_SPECS in src/train_resnet.py.
2. Run:

python src/train_resnet.py