To provide a practical example on how to run the CNN in inference mode, we prepared a short example set of variants extracted from the BLCA-US dataset. Data was preprocessed as described here. In particular, variants were annotated for flanking regions, all variants with a gnomAD record were removed, and per-sample VCF files were merged. Based on the final VCF file BLCA-US_example_20220901.vcf, a .imgb batch was generated.
The CNN can then be run using the following command:
conda activate DeepSom-cnn
python ../../cnn/nn.py --tensor_width=150 --tensor_height=70 --test_dataset=test_imgb.lst --model_weight=../../cnn/models/gnomAD_thr_0/BLCA-US_gnomAD_thr_0_epoch_20_weights_model
The final predictions can then be found in ./predictions/final_predictions.vcf.