NeST-VNN is an interpretable neural network-based model that predicts cell response to a drug. This framework integrates information across multiple levels of cancer cell biology to understand drug response, and can serve to identify and explain biomarkers for clinical application.

Reference:

• Park, S., Silva, E., Singhal, A. et al.,
  A deep learning model of tumor cell architecture elucidates response and resistance to CDK4/6 inhibitors.
  Nat Cancer (2024). https://doi.org/10.1038/s43018-024-00740-1
• Zhao, Singhal, et al.,
  Cancer Mutations Converge on a Collection of Protein Assemblies to Predict Resistance to Replication Stress.
  Cancer Discov 1 March 2024; 14 (3): 508–523. https://doi.org/10.1158/2159-8290.CD-23-0641.

• NeST-VNN Github page

• Module Name: nest_vnn (see the modules page for more information)
• Unusual environment variables set
  • NEST_VNN_HOME  installation directory
  • NEST_VNN_BIN       executable directory
  • NEST_VNN_SRC       source code directory
  • NEST_VNN_DATA  sample data directory

Allocate an interactive session and run the program. Sample session:

[user@biowulf]$ sinteractive --gres=gpu:v100x:1
[user@cn3335 ~]$ module load nest_vnn  
[+] Loading singularity  4.0.1  on cn3335
[+] Loading nest_vnn 20240321

[user@cn3335 ~]$ ls $NEST_VNN_BIN
predict.py  train.py

[user@cn3335 ~]$ train.py -h
...
[user@cn3335 ~]$ predict.py -h 
...

[user@cn3335 ~]$ git clone https://github.com/idekerlab/nest_vnn
[user@cn3335 ~]$ cd nest_vnn/sample
[user@cn3335 ~]$ train.py \
                           -onto ontology.txt \
                           -gene2id gene2ind.txt \
                           -cell2id cell2ind.txt \
                           -train training_data.txt \
                           -mutations cell2mutation.txt \
                           -cn_deletions cell2cndeletion.txt \
                           -cn_amplifications cell2cnamplification.txt \
                           -std ./std.txt \
                           -modeldir ../pretrained_models \
                           -genotype_hiddens 4 \
                           -lr 0.0005 \
                           -cuda 0 \
                           -epoch 50 \
                           -batchsize 64 \
                           -optimize 1 \
                           -zscore_method "auc"

Total number of cell lines = 1244
Total number of genes = 718
There are 1 roots: NEST
There are 131 terms
There are 1 connected componenets
epoch   train_corr      train_loss      true_auc        pred_auc        val_corr        val_loss        grad_norm       elapsed_time
0       -0.0054 40.2999 0.7483  -0.0546 nan     5.0000  0.0016  11.6349
Model saved at epoch 0
1       -0.0088 40.2999 0.7470  -0.0547 nan     5.0000  0.0021  10.2514
2       0.0271  40.2999 0.7496  -0.0548 nan     5.0000  0.0024  9.8969
3       0.0561  40.2999 0.7473  -0.0548 -0.0107 5.0000  0.0037  9.9493
4       0.0125  40.2999 0.7480  -0.0545 0.1610  5.0000  0.0186  9.9294
5       0.0362  40.2996 0.7476  -0.0494 0.2768  4.9999  0.1245  9.8724
6       0.1281  40.2989 0.7473  -0.0364 0.3046  4.9982  0.2250  9.8917
Model saved at epoch 6
7       0.2081  40.2980 0.7488  -0.0210 0.3411  4.9962  0.3514  10.3478
Model saved at epoch 7
8       0.2822  40.2985 0.7483  -0.0042 0.3336  4.9890  0.5787  10.3481
Model saved at epoch 8
9       0.3698  40.2939 0.7485  0.0130  0.3365  4.9884  0.7771  10.3774
10      0.3969  40.2923 0.7491  0.0305  0.3566  4.9842  1.0461  9.9322
Model saved at epoch 10
...
Model saved at epoch 48
49      0.6431  36.6766 0.7482  0.7417  0.2937  4.4424  186.3786        10.4491
[user@cn3335 ~]$ exit
[user@biowulf ~]$