hetnet-ml

Software to quickly extract features from heterogeneous networks for machine learning.

hetnet-ml has been inspired by project Rephetio, this tool attempts to qucikly and efficiently extract metapath based features including the degree-weighted path counts as described by Himmelstein et al..

Installation

Run the following command to install via pip

pip install git+https://github.com/mmayers12/hetnet_ml

Caution!

This repo is currently under development and many features and methods are unstable and subject to change

Usage

See demo folder for further in-depth and up-to-date explanation of features

In [1]: from extractor import MatrixFormattedGraph

In [2]: mg = MatrixFormattedGraph('nodes.csv', 'edges.csv', start_kind='Compound', end_kind='Disease')
Reading file information...
Generating adjcency matrices...
100%|███████████████████████████████████████████████████████████████| 24/24 [01:02<00:00,  6.11s/it]

Weighting matrices by degree with dampening factor 0.4...
100%|███████████████████████████████████████████████████████████████| 25/25 [00:25<00:00,  2.52s/it]

In [3]: res = mg.extract_dwpc(start_nodes='Compound', end_nodes='Disease', n_jobs=32)
Calculating DWPCs...
100%|███████████████████████████████████████████████████████████| 1206/1206 [03:31<00:00,  4.71s/it]

Reformating results...
100%|███████████████████████████████████████████████████████████| 1206/1206 [03:45<00:00,  4.53it/s]

In [4]: res.head(2)
Out[4]:
  compound_id    disease_id  CbGdCuGuD  CuGpCCpGdD  CrCuGaD  CuGeAeGaD  \
0     DB00014  DOID:0050156          0    0.000968        0   0.000425
1     DB00014  DOID:0050425          0    0.000000        0   0.000350

   CtDaGcGaD  CuGuDuGdD  CpDuG<rGuD  CbGbCbGaD     ...       CpDaGaD  \
0          0   0.000662           0          0     ...             0
1          0   0.000000           0          0     ...             0

   CbG<rGcGdD  CdGdDrDrD  CbGuCuGaD  CtDtCrCpD  CuGcGbCtD  CdGiGbCpD  \
0           0          0          0   0.008822          0          0
1           0          0          0   0.001407          0          0

   CdG<rGuCtD  CbGdDaGaD  CuG<rG<rGuD
0           0          0     0.001432
1           0          0     0.000000

[2 rows x 1208 columns]


In [5]: res.shape
Out[5]: (212624, 1208)

In [6]: import bz2

In [7]: with bz2.open('dwpc-features.tsv.bz2', 'wt') as write_file:
   ...:     res.to_csv(write_file, sep='\t', index=False, float_format='%.4g')