aysuvorov/medstats

Collection of scripts and modules for biostatistics and data science simple automation for research process.

Medstats module

Collection of scripts and modules for biostatistics and data science simple automation for research process.

Medstats is a child project of YASP! by Aleksandr Suvorov, senior statistician of Center of analysis of Complex Systems.

Module contents

Describe

Desribe is a module providing various basic statistical functions, plotting, comparative criterial approaches typically used in clinical trials.

Module uses rpy2-module and basic R libraries.

Module is under heavy readjustment.

ModPerf

ModPerf is a module that assesses typical metrics known in medical trials for machine learning algorithms.

Metric functions:

• ModPerf_AUC - computes AUC and AUC 95%CI using bootstrap for binary classifier, probabilities for binary classifier, some quantitative variable;
• ModPerf_Binary - computes confusion matrix and Se, Sp, NPV, PPV and 95%CI using bootstrap for binary classifier output (not probabilities);
• ModPerf_Multiclass - computes confusion matrix and Se, Sp, NPV, PPV and 95%CI using bootstrap for vulticlass classifier output (not probabilities);
• ModPerf_thresholds - computes thresholds for probabilities for binary classifier, some quantitative variable with confusion matrix, Se, Sp, NPV, PPV and 95%CI using bootstrap;

Plotting functions

• ROCPlotter_Binary - plots AUC curve for binary classifier, probabilities for binary classifier, some quantitative variable;
• ROCPlotter_Multiclass - plots AUC curve for multiclass classifier, probabilities for multiclass classifier;

Parenclitic

Parenclitic module involve various functions for network approaches from works of M. Zanin, A. Gorban, A. Zaikin, H. Whitwell, M. Krivonisov, T. Nazarenko.

Various approaches for network analysis:

• parenclitic graphs with a threshold;
• weighted parenclitic graphs;
• synolitic graphs;
• correlation graphs for time series

Common classes

• DataFrameLoader - common class for loading data (for all types of approaches);
• Prct - common class for parenclytic approach;
• Snltc - common class for synolytic approach;
• Corr - common class for correlation approach;

Common functions

• graph_plotter - plots single weighted graph;
• chars -computes various characteristics of a single graph;

PhisioPatterns

• test module for time series and longitudinal data (ECG, EEG). Under heavy development;

Test platform

• MLSelectionFlow_devel - programm for automatic feature selection (SHAP) and model selection (optuna module);