pySAR works mainly via JSON configuration files. There are many different customisable parameters for the functionalities in pySAR including the metaparameters of some of the available protein descriptors, all Digital Signal Processing (DSP) parameters in the pyDSP module, the type of regression model to use and parameters specific to the dataset - a description of each parameter is available in the example below.
These config files offer a more straightforward way of making any changes to the pySAR pipeline. The names of All the parameters as listed in the example config files must remain unchanged, only the value of each parameter should be changed, any parameters not being used can be set to null. Additionally, you can pass in the individual parameter names and values to the pySAR and Encoding classes when numerically encoding the protein sequences via kwargs. An example of the config file used in my research project (thermostability.json), with all of the available parameters, can be seen below.
{
"dataset":
{
"dataset": "thermostability.txt",
"sequence_col": "sequence",
"activity": "t50"
},
"model":
{
"algorithm": "plsregression",
"parameters": "",
"test_split": 0.2
},
"descriptors":
{
"descriptors_csv": "descriptors_thermostability.csv",
"moreaubroto_autocorrelation":
{
"lag": 30,
"properties": ["CIDH920105", "BHAR880101", "CHAM820101", "CHAM820102",
"CHOC760101", "BIGC670101", "CHAM810101", "DAYM780201"],
"normalize": 1
},
"moran_autocorrelation":
{
"lag": 30,
"properties": ["CIDH920105", "BHAR880101", "CHAM820101", "CHAM820102",
"CHOC760101", "BIGC670101", "CHAM810101", "DAYM780201"],
"normalize": 1
},
"geary_autocorrelation":
{
"lag": 30,
"properties": ["CIDH920105", "BHAR880101", "CHAM820101", "CHAM820102",
"CHOC760101", "BIGC670101", "CHAM810101", "DAYM780201"],
"normalize": 1
},
"ctd":
{
"property": "hydrophobicity",
"all": 1
},
"sequence_order_coupling_number":
{
"lag": 30,
"distance_matrix": "schneider-wrede.json"
},
"quasi_sequence_order":
{
"lag": 30,
"weight": 0.1,
"distance_matrix": "schneider-wrede.json"
},
"pseudo_amino_acid_composition":
{
"lambda": 30,
"weight": 0.05,
"properties": []
},
"amphiphilic_pseudo_amino_acid_composition":
{
"lambda": 30,
"weight": 0.5
}
},
"pyDSP":
{
"use_dsp": 0,
"spectrum": "power",
"window": {
"type": "hamming",
"sym": 1,
"beta": null,
"alpha": null,
"nbar": null,
"sll": null,
"norm": null
},
"filter": {
"type": null,
"window_length": 5,
"polyorder": 2,
"deriv": 0,
"delta": 1,
"mode": "interp"
}
}
}Dataset Parameters:
dataset[dataset]- path of dataset.dataset[sequence_col]- name of sequence column in dataset holding protein sequences, if left blank 'sequence' will be used by default.dataset[activity]- name of protein activity column in dataset being studied.
Model Parameters:
model[algorithm]- name of ML regression algorithm being utilised.model[parameters]- parameters of ML regression algorithm being utilised. If left blank, the algorithm's default parameters will be used.model[test_split]- training/test split for training the ML regression model, between 0 and 1. If left blank, a default split of 0.2 is used.
Descriptor Parameters:
-
descriptors[descriptors_csv]- path to csv file of pre-calculated descriptor values of a dataset, saves time having to recalculate the features each time. -
descriptors[moreaubroto_autocorrelation][lag] / descriptors[moran_autocorrelation][lag] / descriptors[geary_autocorrelation][lag]- The maximum lag value for each of the autocorrelation descriptors. If invalid value input then a default of 30 is used. -
descriptors[moreaubroto_autocorrelation][properties] / descriptors[moran_autocorrelation][properties] / descriptors[geary_autocorrelation][properties]- List of protein physiochemical and structural descriptors used in the calculation of each of the autocorrelation descriptors, properties must be a list of their AAIndex number/accession numbers. There must be a least 1 property value input. -
descriptors[moreaubroto_autocorrelation][normalize] / descriptors[moran_autocorrelation][normalize] / descriptors[geary_autocorrelation][normalize]- rescale/normalize Autocorrelation values into range of 0-1. -
descriptors[ctd][property]- list of 1 or more physiochemical properties to use when calculating CTD descriptors. List of available input properties: hydrophobicity, normalized_vdwv, polarity, charge, secondary_struct, solvent_accessibility, polarizability. If no properties input then hydrophobicity used by default. -
descriptors[ctd][all]- if True then all 7 of the available physiochemical descriptors will be used when calculating the CTD descriptors. Each proeprty generates 21 features so using all properties will output 147 features. Only 1 property used by default. -
descriptors[sequence_order_coupling_number][maxlag]- maximum lag; length of the protein must be not less than maxlag. -
descriptors[sequence_order_coupling_number][distance_matrix]- physiochemical distance matrix name for calculating sequence order coupling number. -
descriptors[quasi_sequence_order][maxlag]- maximum lag; length of the protein must be not less than maxlag. -
descriptors[quasi_sequence_order][weight]- weighting factor to use when calculating descriptor. -
descriptors[quasi_sequence_order][distance_matrix]- physiochemical distance matrix name for calculating quasi sequence order. -
descriptors[pseudo_amino_acid_composition][lambda]- lambda parameter that reflects the rank correlation and should be a non-negative integer and not larger than the length of the protein sequence. -
descriptors[pseudo_amino_acid_composition][weight]- weighting factor to use when calculating descriptor. -
descriptors[pseudo_amino_acid_composition][properties]- 1 or more amino acid index properties from the AAI database used for calculating the sequence-order. -
descriptors[amphiphilic_pseudo_amino_acid_composition][lambda]- lambda parameter that reflects the rank correlation and should be a non-negative integer and not larger than the length of the protein sequence. -
descriptors[amphiphilic_pseudo_amino_acid_composition][weight]- weighting factor to use when calculating descriptor.
DSP Parameters:
pyDSP[use_dsp]- whether or not to apply Digital Signal Processing (DSP) techniques to the features passed into the model. If true, the values of the next DSP parameters will be applied to the features.pyDSP[spectrum]- which frequency output/informational spectra to use from the generated types of signals from DSP to use e.g power, absolute, imaginery, real.pyDSP[window][type]- convolutional window to apply to the signal output, pySAR supports: hamming, blackman, blackmanharris, gaussian, bartlett, kaiser, barthann, bohman, chebwin, cosine, exponential, flattop, hann, boxcar, hanning, nuttall, parzen, triang and tukey.pyDSP[filter][type]- window filter to apply to the signal output, pySAR supports: savgol, medfilt, symiirorder1, lfilter and hilbert.