deep-geometric-pathway

Genomic graph-learning LOG

March 23rd

1. Implement regression_hops_sampler() function. The function can return an always $1$ batch size object loader. The return serves for the regression training without considering the geometric structure, but each training set of loader does from the num_hops. So, it is kind of fair for further geometric pathway training. (DONE)
2. Normalized the data, fixed the possible $NAN$ issue
3. Implemented the pathway_LassoCV.py to compute the regression, all the plot results are shown as GROUND TRUTH vs PREDICTION. They are in ./results/Lasso folder

Example:

See detail is ./regression_hop_sampler.ipynb ,./pathway_LassoCV.ipynb. Also some results figures: ['ABL1'].pdf ['APAF1'].pdf

March 22nd

1. Implement cancer_data() function, which operates the Breast Caner data in ./BreastCancer/Data_RNASeq2.mat (DONE)
2. Implement hops_sampler() function that sample the overall genomeic/pathway info and return the dataloader in specific batch_size required. (DONE)

Example:

The detailed example is available in ./test_notebook/ and ./sampler_test.ipynb

March 21st

1. The pure text raw data ./GenomicData/Gene_DATA/sourcePathway.txt processed method (DONE)
2. The utility method data_fetch() class has been DONE
3. data_fetch is able to exclude the node type that is no necessary wanted

Example:

$(LIST) is optional input list type. e.g., ['protein', 'abstract']

from utils.data_fetch as dffter
dffter = data_fetch(argv_list=$(LIST))

March 12th

1. Have more clear idea of dataset
2. Get the node information and pattern information
3. Learned Regular expression(Regex)
4. Convert rough data from .mat to .txt for future purpose
5. The link and node info are in ./GenomicData/Gene_DATA/sourcePathway.txt
6. The further genomic activity feature is stored in ./GenomicData/Data_RNASeq2.mat NEED TO PROCESS LATER

The following is the regex pattern that can make extract information from source file really quick and efficient.

// General node pattern
^([a-z]+)\s([\w\/\-()+]+)$

// Link pattern extraction
^([\w\/\-()+]+)\s([\w\/\-()+]+)\s([\w\>\|-]+)$